Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.13643 · 2026-10-09
Key numbers: 14 tasks
Signals: closed-loop success reported · real-robot experiments
Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.13643 · 2026-10-09\nKey numbers: 14 tasks\nSignals: closed-loop success reported · real-robot experiments\n\nAutomating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation.\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation.", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "arXiv 2610.13643" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "arxiv:2610.13643", "url": "https://arxiv.org/abs/2610.13643", "title": "AutoBio+: Benchmarking Dexterous Manipulation with Dynamic Experimental Phenomena in Realistic Laboratories", "publisher": "arXiv 2610.13643", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-09", "radar_published": "2026-10-12T23:40:00Z", "numbers": [ "14 tasks" ], "signals": [ "closed_loop", "real_robot" ], "excerpt": "Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.13117", "url": "https://arxiv.org/abs/2610.13117", "title": "Forward Dynamics Models", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.13117 · 2026-10-09
Key numbers: 99.4% · 13%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.13117 · 2026-10-09\nKey numbers: 99.4% · 13%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nVision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively.\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively.", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "arXiv 2610.13117" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "arxiv:2610.13117", "url": "https://arxiv.org/abs/2610.13117", "title": "Forward Dynamics Models", "publisher": "arXiv 2610.13117", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-09", "radar_published": "2026-10-12T23:40:00Z", "numbers": [ "99.4%", "13%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Vision-language-action (VLA) models and world action models (WAMs) have advanced robotic policy learning by building on pretrained vision-language models (VLMs) and video generative models (VGMs), respectively." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.13345", "url": "https://arxiv.org/abs/2610.13345", "title": "Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.13345 · 2026-10-09
Key numbers: 1,000 hours
Signals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights
Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.13345 · 2026-10-09\nKey numbers: 1,000 hours\nSignals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights\n\nExisting real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts.\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts.", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "arXiv 2610.13345" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "arxiv:2610.13345", "url": "https://arxiv.org/abs/2610.13345", "title": "Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience", "publisher": "arXiv 2610.13345", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-10-09", "radar_published": "2026-10-12T23:40:00Z", "numbers": [ "1,000 hours" ], "signals": [ "numbers", "closed_loop", "real_robot", "open_release" ], "excerpt": "Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:aws-physical-ai:a07068e67300", "url": "https://aws.amazon.com/blogs/physical-ai/3d-gaussian-splatting-in-practice-industry-use-cases/", "title": "3D Gaussian Splatting in practice: Industry use cases", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [O]
Source: AWS Physical AI Blog · 2026-10-06
Gaussian Splatting is dramatically lowering the cost and complexity of photorealistic 3D, making once impractical use cases like near real time capture, robotics, and simulation viable across media, construction, manufacturing, and real…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [O]\nSource: AWS Physical AI Blog · 2026-10-06\n\nGaussian Splatting is dramatically lowering the cost and complexity of photorealistic 3D, making once impractical use cases like near real time capture, robotics, and simulation viable across media, construction, manufacturing, and real…\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "Gaussian Splatting is dramatically lowering the cost and complexity of photorealistic 3D, making once impractical use cases like near real time capture, robotics, and simulation viable across media, construction, manufacturing, and real…", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "AWS Physical AI Blog" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[O]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "aws-physical-ai:a07068e67300", "url": "https://aws.amazon.com/blogs/physical-ai/3d-gaussian-splatting-in-practice-industry-use-cases/", "title": "3D Gaussian Splatting in practice: Industry use cases", "publisher": "AWS Physical AI Blog", "source_id": "aws-physical-ai", "evidence": "O", "lane": "simeval", "published": "2026-10-06", "radar_published": "2026-10-12T23:40:00Z", "numbers": [], "signals": [], "excerpt": "Gaussian Splatting is dramatically lowering the cost and complexity of photorealistic 3D, making once impractical use cases like near real time capture, robotics, and simulation viable across media, construction, manufacturing, and real…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:fb73046bfc9e", "url": "https://www.therobotreport.com/closing-reality-gap-why-real-time-execution-has-become-new-competitive-advantage/", "title": "Closing the reality gap: Why real-time execution has become the new competitive advantage", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [M]
Source: The Robot Report · 2026-10-12
Signals: latency or control-rate numbers
Daifuku's Cecile Dick-Calmes said success is no longer measured solely by automation density or equipment utilization.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [M]\nSource: The Robot Report · 2026-10-12\nSignals: latency or control-rate numbers\n\nDaifuku's Cecile Dick-Calmes said success is no longer measured solely by automation density or equipment utilization.\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "Daifuku's Cecile Dick-Calmes said success is no longer measured solely by automation density or equipment utilization.", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "the-robot-report:fb73046bfc9e", "url": "https://www.therobotreport.com/closing-reality-gap-why-real-time-execution-has-become-new-competitive-advantage/", "title": "Closing the reality gap: Why real-time execution has become the new competitive advantage", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "edge", "published": "2026-10-12", "radar_published": "2026-10-12T23:40:00Z", "numbers": [], "signals": [ "latency" ], "excerpt": "Daifuku's Cecile Dick-Calmes said success is no longer measured solely by automation density or equipment utilization." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:9c414b247031", "url": "https://www.leiphone.com/category/transportation/Y6GLcnioywVn3u7K.html", "title": "费莫一刘念邱：智驾无图的经验，也能迁移到机器人 | 物理AI 50人", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-12
刘念邱很难被归进典型的具身创业者模板里。 他不是从机器人实验室出来的学者，也不是过去两年才进入 Physical AI 的连续创业者。\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-12\n\n刘念邱很难被归进典型的具身创业者模板里。 他不是从机器人实验室出来的学者，也不是过去两年才进入 Physical AI 的连续创业者。\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "刘念邱很难被归进典型的具身创业者模板里。 他不是从机器人实验室出来的学者，也不是过去两年才进入 Physical AI 的连续创业者。", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "leiphone:9c414b247031", "url": "https://www.leiphone.com/category/transportation/Y6GLcnioywVn3u7K.html", "title": "费莫一刘念邱：智驾无图的经验，也能迁移到机器人 | 物理AI 50人", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "hardware", "published": "2026-10-12", "radar_published": "2026-10-12T23:40:00Z", "numbers": [], "signals": [], "excerpt": "刘念邱很难被归进典型的具身创业者模板里。 他不是从机器人实验室出来的学者，也不是过去两年才进入 Physical AI 的连续创业者。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:28b795de6185", "url": "https://www.leiphone.com/category/industrynews/ZFAMCctM7KUw5gr3.html", "title": "产业出题，科技答题：2026 IEIC国际具身智能技能大赛，用真实岗位任务考核机器人落地能力", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-12
实景比拼赋能科创，助力人形机器人技术规模化落地！ 编辑丨李希 为深入贯彻落实工业和信息化部、国务院国资委《 2026年度人形机器人与具身智能实景实训专项行动》部署，加快推动我国机器人与具身智能技术创新突破与规模化落地，国际具身智能技能大赛检阅赛将于2026年10月23日至25日在上海举行。\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-12\n\n实景比拼赋能科创，助力人形机器人技术规模化落地！ 编辑丨李希 为深入贯彻落实工业和信息化部、国务院国资委《 2026年度人形机器人与具身智能实景实训专项行动》部署，加快推动我国机器人与具身智能技术创新突破与规模化落地，国际具身智能技能大赛检阅赛将于2026年10月23日至25日在上海举行。\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "实景比拼赋能科创，助力人形机器人技术规模化落地！ 编辑丨李希 为深入贯彻落实工业和信息化部、国务院国资委《 2026年度人形机器人与具身智能实景实训专项行动》部署，加快推动我国机器人与具身智能技术创新突破与规模化落地，国际具身智能技能大赛检阅赛将于2026年10月23日至25日在上海举行。", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Data engine (collection / curation / scaling)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "leiphone:28b795de6185", "url": "https://www.leiphone.com/category/industrynews/ZFAMCctM7KUw5gr3.html", "title": "产业出题，科技答题：2026 IEIC国际具身智能技能大赛，用真实岗位任务考核机器人落地能力", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "data", "published": "2026-10-12", "radar_published": "2026-10-12T23:40:00Z", "numbers": [], "signals": [], "excerpt": "实景比拼赋能科创，助力人形机器人技术规模化落地！ 编辑丨李希 为深入贯彻落实工业和信息化部、国务院国资委《 2026年度人形机器人与具身智能实景实训专项行动》部署，加快推动我国机器人与具身智能技术创新突破与规模化落地，国际具身智能技能大赛检阅赛将于2026年10月23日至25日在上海举行。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:9be94d2c68ce", "url": "https://www.leiphone.com/category/industrynews/ukm5yjK3oT5X29yV.html", "title": "百度世界2026将于11月12日在北京举办，主题为「下一幕，智能组织」", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-12
10月12日，百度官方宣布，百度世界2026将于11月12日在北京·国家会议中心二期举办。目前，大会官网已开启售票通道（https://baiduworld2026.events.net.cn/tickets）。\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-12\n\n10月12日，百度官方宣布，百度世界2026将于11月12日在北京·国家会议中心二期举办。目前，大会官网已开启售票通道（https://baiduworld2026.events.net.cn/tickets）。\n\nFull radar for 2026-10-13: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-13.en.html", "summary": "10月12日，百度官方宣布，百度世界2026将于11月12日在北京·国家会议中心二期举办。目前，大会官网已开启售票通道（https://baiduworld2026.events.net.cn/tickets）。", "date_published": "2026-10-12T23:40:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-13", "id": "leiphone:9be94d2c68ce", "url": "https://www.leiphone.com/category/industrynews/ukm5yjK3oT5X29yV.html", "title": "百度世界2026将于11月12日在北京举办，主题为「下一幕，智能组织」", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "training", "published": "2026-10-12", "radar_published": "2026-10-12T23:40:00Z", "numbers": [], "signals": [], "excerpt": "10月12日，百度官方宣布，百度世界2026将于11月12日在北京·国家会议中心二期举办。目前，大会官网已开启售票通道（https://baiduworld2026.events.net.cn/tickets）。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12368", "url": "https://arxiv.org/abs/2610.12368", "title": "LiteNWM: Efficient Latent World Models for Onboard Visual Navigation in the Wild", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.12368 · 2026-10-08
Key numbers: 17.56% · 16.2% · 43.3% · 83.3%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Direct visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.12368 · 2026-10-08\nKey numbers: 17.56% · 16.2% · 43.3% · 83.3%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nDirect visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Direct visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12368" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12368", "url": "https://arxiv.org/abs/2610.12368", "title": "LiteNWM: Efficient Latent World Models for Onboard Visual Navigation in the Wild", "publisher": "arXiv 2610.12368", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [ "17.56%", "16.2%", "43.3%", "83.3%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Direct visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11480", "url": "https://arxiv.org/abs/2610.11480", "title": "RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.11480 · 2026-10-08
Key numbers: 100 tasks · 77.0% · 90.0% · 73.8%
Signals: quantitative results · closed-loop success reported
Embodied coding agents can combine modular robot skills with frozen end-to-end policies, yet effective composition requires anticipating which policy family will succeed in the current physical state.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.11480 · 2026-10-08\nKey numbers: 100 tasks · 77.0% · 90.0% · 73.8%\nSignals: quantitative results · closed-loop success reported\n\nEmbodied coding agents can combine modular robot skills with frozen end-to-end policies, yet effective composition requires anticipating which policy family will succeed in the current physical state.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Embodied coding agents can combine modular robot skills with frozen end-to-end policies, yet effective composition requires anticipating which policy family will succeed in the current physical state.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.11480" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.11480", "url": "https://arxiv.org/abs/2610.11480", "title": "RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes", "publisher": "arXiv 2610.11480", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [ "100 tasks", "77.0%", "90.0%", "73.8%" ], "signals": [ "numbers", "closed_loop" ], "excerpt": "Embodied coding agents can combine modular robot skills with frozen end-to-end policies, yet effective composition requires anticipating which policy family will succeed in the current physical state." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12440", "url": "https://arxiv.org/abs/2610.12440", "title": "Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [R]
Source: arXiv 2610.12440 · 2026-10-08
Signals: closed-loop success reported
Human demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots.\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [R]\nSource: arXiv 2610.12440 · 2026-10-08\nSignals: closed-loop success reported\n\nHuman demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Human demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12440" } ], "tags": [ "Data engine (collection / curation / scaling)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12440", "url": "https://arxiv.org/abs/2610.12440", "title": "Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation", "publisher": "arXiv 2610.12440", "source_id": "arxiv", "evidence": "R", "lane": "data", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "Human demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12407", "url": "https://arxiv.org/abs/2610.12407", "title": "LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.12407 · 2026-10-08
Signals: closed-loop success reported
World action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.12407 · 2026-10-08\nSignals: closed-loop success reported\n\nWorld action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "World action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12407" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12407", "url": "https://arxiv.org/abs/2610.12407", "title": "LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC", "publisher": "arXiv 2610.12407", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "World action models (WAMs) predict actions and future observations, typically from a reconstruction-based representation that carries noisy, redundant information which can complicate downstream predictions." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12457", "url": "https://arxiv.org/abs/2610.12457", "title": "SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic Manipulation", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.12457 · 2026-10-08
Key numbers: 26.7% · 66.7% · 0% · 100%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Frontier multimodal foundation models (e.g., GPT-6 Astra) have recently shown strong potential for direct robotic control, yet their performance on fine manipulation remains limited.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.12457 · 2026-10-08\nKey numbers: 26.7% · 66.7% · 0% · 100%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nFrontier multimodal foundation models (e.g., GPT-6 Astra) have recently shown strong potential for direct robotic control, yet their performance on fine manipulation remains limited.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Frontier multimodal foundation models (e.g., GPT-6 Astra) have recently shown strong potential for direct robotic control, yet their performance on fine manipulation remains limited.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12457" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12457", "url": "https://arxiv.org/abs/2610.12457", "title": "SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic Manipulation", "publisher": "arXiv 2610.12457", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [ "26.7%", "66.7%", "0%", "100%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Frontier multimodal foundation models (e.g., GPT-6 Astra) have recently shown strong potential for direct robotic control, yet their performance on fine manipulation remains limited." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12249", "url": "https://arxiv.org/abs/2610.12249", "title": "Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.12249 · 2026-10-08
Signals: closed-loop success reported · latency or control-rate numbers
We study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.12249 · 2026-10-08\nSignals: closed-loop success reported · latency or control-rate numbers\n\nWe study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "We study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12249" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12249", "url": "https://arxiv.org/abs/2610.12249", "title": "Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods", "publisher": "arXiv 2610.12249", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "latency" ], "excerpt": "We study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11809", "url": "https://arxiv.org/abs/2610.11809", "title": "WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.11809 · 2026-10-08
Signals: closed-loop success reported · latency or control-rate numbers
Robust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision…\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.11809 · 2026-10-08\nSignals: closed-loop success reported · latency or control-rate numbers\n\nRobust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision…\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Robust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision…", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.11809" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.11809", "url": "https://arxiv.org/abs/2610.11809", "title": "WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors", "publisher": "arXiv 2610.11809", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "latency" ], "excerpt": "Robust navigation in cluttered environments remains a fundamental challenge for quadrotors, particularly when strong wind disturbances arise, which perturb vehicle dynamics, limit control authority, and substantially increase collision…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12467", "url": "https://arxiv.org/abs/2610.12467", "title": "CSF: Contextual Safety Filtering for Motion Generators", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [R]
Source: arXiv 2610.12467 · 2026-10-08
Key numbers: 90% · 100%
Signals: quantitative results · real-robot experiments
Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person.\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [R]\nSource: arXiv 2610.12467 · 2026-10-08\nKey numbers: 90% · 100%\nSignals: quantitative results · real-robot experiments\n\nText-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person.\n\nFull radar for 2026-10-12: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-12.en.html", "summary": "Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person.", "date_published": "2026-10-11T23:40:00Z", "authors": [ { "name": "arXiv 2610.12467" } ], "tags": [ "Safety, permissions & compliance", "[R]" ], "language": "en", "_radar": { "date": "2026-10-12", "id": "arxiv:2610.12467", "url": "https://arxiv.org/abs/2610.12467", "title": "CSF: Contextual Safety Filtering for Motion Generators", "publisher": "arXiv 2610.12467", "source_id": "arxiv", "evidence": "R", "lane": "safety", "published": "2026-10-08", "radar_published": "2026-10-11T23:40:00Z", "numbers": [ "90%", "100%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Text-conditioned motion generators produce trackable whole-body motion, but they have no notion of scene-dependent safety: the same action may target an object or a person." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12185", "url": "https://arxiv.org/abs/2610.12185", "title": "RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.12185 · 2026-10-08
Key numbers: 8.70% · 6.28% · 6.83% · 74.04%
Signals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers
Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.12185 · 2026-10-08\nKey numbers: 8.70% · 6.28% · 6.83% · 74.04%\nSignals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers\n\nReliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.12185" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.12185", "url": "https://arxiv.org/abs/2610.12185", "title": "RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control", "publisher": "arXiv 2610.12185", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [ "8.70%", "6.28%", "6.83%", "74.04%" ], "signals": [ "numbers", "real_robot", "open_release", "latency" ], "excerpt": "Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11283", "url": "https://arxiv.org/abs/2610.11283", "title": "Being-M0.7: A Latent World-Action Model for Humanoid Robots", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [R]
Source: arXiv 2610.11283 · 2026-10-08
Key numbers: 10,000 hours
Signals: quantitative results · closed-loop success reported · real-robot experiments
Humanoid loco-manipulation requires coordinated locomotion and manipulation informed by future scene evolution and whole-body motion, yet learning these capabilities is constrained by scarce robot demonstrations.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [R]\nSource: arXiv 2610.11283 · 2026-10-08\nKey numbers: 10,000 hours\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nHumanoid loco-manipulation requires coordinated locomotion and manipulation informed by future scene evolution and whole-body motion, yet learning these capabilities is constrained by scarce robot demonstrations.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Humanoid loco-manipulation requires coordinated locomotion and manipulation informed by future scene evolution and whole-body motion, yet learning these capabilities is constrained by scarce robot demonstrations.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.11283" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.11283", "url": "https://arxiv.org/abs/2610.11283", "title": "Being-M0.7: A Latent World-Action Model for Humanoid Robots", "publisher": "arXiv 2610.11283", "source_id": "arxiv", "evidence": "R", "lane": "hardware", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [ "10,000 hours" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Humanoid loco-manipulation requires coordinated locomotion and manipulation informed by future scene evolution and whole-body motion, yet learning these capabilities is constrained by scarce robot demonstrations." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12231", "url": "https://arxiv.org/abs/2610.12231", "title": "Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.12231 · 2026-10-08
Signals: closed-loop success reported · real-robot experiments
Learning from demonstration has enabled impressive robot behaviors. A common choice for policy learning is to use diffusion or flow matching (Flow-Policies), which often outperforms direct action regression trained with mean squared error…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.12231 · 2026-10-08\nSignals: closed-loop success reported · real-robot experiments\n\nLearning from demonstration has enabled impressive robot behaviors. A common choice for policy learning is to use diffusion or flow matching (Flow-Policies), which often outperforms direct action regression trained with mean squared error…\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Learning from demonstration has enabled impressive robot behaviors. A common choice for policy learning is to use diffusion or flow matching (Flow-Policies), which often outperforms direct action regression trained with mean squared error…", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.12231" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.12231", "url": "https://arxiv.org/abs/2610.12231", "title": "Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning", "publisher": "arXiv 2610.12231", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot" ], "excerpt": "Learning from demonstration has enabled impressive robot behaviors. A common choice for policy learning is to use diffusion or flow matching (Flow-Policies), which often outperforms direct action regression trained with mean squared error…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12386", "url": "https://arxiv.org/abs/2610.12386", "title": "ARC: A Reasoning Recipe for Robot Foundation Models", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.12386 · 2026-10-08
Signals: closed-loop success reported · real-robot experiments
The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.12386 · 2026-10-08\nSignals: closed-loop success reported · real-robot experiments\n\nThe prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.12386" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.12386", "url": "https://arxiv.org/abs/2610.12386", "title": "ARC: A Reasoning Recipe for Robot Foundation Models", "publisher": "arXiv 2610.12386", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot" ], "excerpt": "The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12046", "url": "https://arxiv.org/abs/2610.12046", "title": "SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [R]
Source: arXiv 2610.12046 · 2026-10-08
Key numbers: 27% · 65% · 87%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Dexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes.\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [R]\nSource: arXiv 2610.12046 · 2026-10-08\nKey numbers: 27% · 65% · 87%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nDexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Dexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.12046" } ], "tags": [ "Data engine (collection / curation / scaling)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.12046", "url": "https://arxiv.org/abs/2610.12046", "title": "SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills", "publisher": "arXiv 2610.12046", "source_id": "arxiv", "evidence": "R", "lane": "data", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [ "27%", "65%", "87%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Dexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12432", "url": "https://arxiv.org/abs/2610.12432", "title": "FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [R]
Source: arXiv 2610.12432 · 2026-10-08
Key numbers: 99.95% · 97%
Signals: quantitative results · real-robot experiments
Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates.\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [R]\nSource: arXiv 2610.12432 · 2026-10-08\nKey numbers: 99.95% · 97%\nSignals: quantitative results · real-robot experiments\n\nSafe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.12432" } ], "tags": [ "Safety, permissions & compliance", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.12432", "url": "https://arxiv.org/abs/2610.12432", "title": "FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems", "publisher": "arXiv 2610.12432", "source_id": "arxiv", "evidence": "R", "lane": "safety", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [ "99.95%", "97%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11416", "url": "https://arxiv.org/abs/2610.11416", "title": "Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.11416 · 2026-10-08
Signals: real-robot experiments
Vision-Language-Action (VLA) models have emerged as a prominent framework for complex robotic manipulation, building on the strong semantic understanding of pretrained Vision-Language Models (VLMs).\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.11416 · 2026-10-08\nSignals: real-robot experiments\n\nVision-Language-Action (VLA) models have emerged as a prominent framework for complex robotic manipulation, building on the strong semantic understanding of pretrained Vision-Language Models (VLMs).\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "Vision-Language-Action (VLA) models have emerged as a prominent framework for complex robotic manipulation, building on the strong semantic understanding of pretrained Vision-Language Models (VLMs).", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "arXiv 2610.11416" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "arxiv:2610.11416", "url": "https://arxiv.org/abs/2610.11416", "title": "Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer", "publisher": "arXiv 2610.11416", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-10T23:40:00Z", "numbers": [], "signals": [ "real_robot" ], "excerpt": "Vision-Language-Action (VLA) models have emerged as a prominent framework for complex robotic manipulation, building on the strong semantic understanding of pretrained Vision-Language Models (VLMs)." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:0698575e496b", "url": "https://www.therobotreport.com/why-humanoid-robot-demos-still-fail-the-generalization-test/", "title": "Why humanoid robot demos still fail the generalization test", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: The Robot Report · 2026-10-09
Signals: real-robot experiments
How better data and human judgment can help robots move beyond demos toward reliable real-world performance.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: The Robot Report · 2026-10-09\nSignals: real-robot experiments\n\nHow better data and human judgment can help robots move beyond demos toward reliable real-world performance.\n\nFull radar for 2026-10-11: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-11.en.html", "summary": "How better data and human judgment can help robots move beyond demos toward reliable real-world performance.", "date_published": "2026-10-10T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-11", "id": "the-robot-report:0698575e496b", "url": "https://www.therobotreport.com/why-humanoid-robot-demos-still-fail-the-generalization-test/", "title": "Why humanoid robot demos still fail the generalization test", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "hardware", "published": "2026-10-09", "radar_published": "2026-10-10T23:40:00Z", "numbers": [], "signals": [ "real_robot" ], "excerpt": "How better data and human judgment can help robots move beyond demos toward reliable real-world performance." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.09254", "url": "https://arxiv.org/abs/2610.09254", "title": "RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.09254 · 2026-10-07
Key numbers: 71% · 7.1x · 3.6x
Signals: quantitative results · closed-loop success reported · real-robot experiments
Simulation enables large-scale, low-cost robot data generation, but policies trained in simulation often fail to transfer to the real world due to the sim-to-real visual discrepancies.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.09254 · 2026-10-07\nKey numbers: 71% · 7.1x · 3.6x\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nSimulation enables large-scale, low-cost robot data generation, but policies trained in simulation often fail to transfer to the real world due to the sim-to-real visual discrepancies.\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "Simulation enables large-scale, low-cost robot data generation, but policies trained in simulation often fail to transfer to the real world due to the sim-to-real visual discrepancies.", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "arXiv 2610.09254" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "arxiv:2610.09254", "url": "https://arxiv.org/abs/2610.09254", "title": "RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer", "publisher": "arXiv 2610.09254", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-07", "radar_published": "2026-10-10T07:15:00Z", "numbers": [ "71%", "7.1x", "3.6x" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Simulation enables large-scale, low-cost robot data generation, but policies trained in simulation often fail to transfer to the real world due to the sim-to-real visual discrepancies." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11322", "url": "https://arxiv.org/abs/2610.11322", "title": "USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for Simulation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.11322 · 2026-10-08
Signals: real-robot experiments
Geometrically faithful and functional articulated 3D assets are essential for real-to-sim robot manipulation, where policies trained in simulation must transfer to physical objects.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.11322 · 2026-10-08\nSignals: real-robot experiments\n\nGeometrically faithful and functional articulated 3D assets are essential for real-to-sim robot manipulation, where policies trained in simulation must transfer to physical objects.\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "Geometrically faithful and functional articulated 3D assets are essential for real-to-sim robot manipulation, where policies trained in simulation must transfer to physical objects.", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "arXiv 2610.11322" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "arxiv:2610.11322", "url": "https://arxiv.org/abs/2610.11322", "title": "USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for Simulation", "publisher": "arXiv 2610.11322", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-10T07:15:00Z", "numbers": [], "signals": [ "real_robot" ], "excerpt": "Geometrically faithful and functional articulated 3D assets are essential for real-to-sim robot manipulation, where policies trained in simulation must transfer to physical objects." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12026", "url": "https://arxiv.org/abs/2610.12026", "title": "Humanoid World Action Model With Joint State--Action Generation", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.12026 · 2026-10-08
Key numbers: 70.6% · 43.3%
Signals: quantitative results · closed-loop success reported
Humanoid robots are a promising platform for general-purpose manipulation. Recent Vision-Language-Action (VLA) policies learn actions directly from multimodal observations, while World Action Models (WAMs) further incorporate future…\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.12026 · 2026-10-08\nKey numbers: 70.6% · 43.3%\nSignals: quantitative results · closed-loop success reported\n\nHumanoid robots are a promising platform for general-purpose manipulation. Recent Vision-Language-Action (VLA) policies learn actions directly from multimodal observations, while World Action Models (WAMs) further incorporate future…\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "Humanoid robots are a promising platform for general-purpose manipulation. Recent Vision-Language-Action (VLA) policies learn actions directly from multimodal observations, while World Action Models (WAMs) further incorporate future…", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "arXiv 2610.12026" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "arxiv:2610.12026", "url": "https://arxiv.org/abs/2610.12026", "title": "Humanoid World Action Model With Joint State--Action Generation", "publisher": "arXiv 2610.12026", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-10T07:15:00Z", "numbers": [ "70.6%", "43.3%" ], "signals": [ "numbers", "closed_loop" ], "excerpt": "Humanoid robots are a promising platform for general-purpose manipulation. Recent Vision-Language-Action (VLA) policies learn actions directly from multimodal observations, while World Action Models (WAMs) further incorporate future…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12089", "url": "https://arxiv.org/abs/2610.12089", "title": "ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks", "content_html": "
Lane: Systems & orchestration (long-horizon / multi-policy)
Evidence: [R]
Source: arXiv 2610.12089 · 2026-10-08
Key numbers: 56% · 78.7% · 49.3% · 4.0%
Signals: quantitative results · closed-loop success reported
Long-horizon mobile manipulation requires a robot to navigate multi-room environments and execute a sequence of manipulation skills under a single natural language instruction.\n", "content_text": "Lane: Systems & orchestration (long-horizon / multi-policy)\nEvidence: [R]\nSource: arXiv 2610.12089 · 2026-10-08\nKey numbers: 56% · 78.7% · 49.3% · 4.0%\nSignals: quantitative results · closed-loop success reported\n\nLong-horizon mobile manipulation requires a robot to navigate multi-room environments and execute a sequence of manipulation skills under a single natural language instruction.\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "Long-horizon mobile manipulation requires a robot to navigate multi-room environments and execute a sequence of manipulation skills under a single natural language instruction.", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "arXiv 2610.12089" } ], "tags": [ "Systems & orchestration (long-horizon / multi-policy)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "arxiv:2610.12089", "url": "https://arxiv.org/abs/2610.12089", "title": "ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks", "publisher": "arXiv 2610.12089", "source_id": "arxiv", "evidence": "R", "lane": "systems", "published": "2026-10-08", "radar_published": "2026-10-10T07:15:00Z", "numbers": [ "56%", "78.7%", "49.3%", "4.0%" ], "signals": [ "numbers", "closed_loop" ], "excerpt": "Long-horizon mobile manipulation requires a robot to navigate multi-room environments and execute a sequence of manipulation skills under a single natural language instruction." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:ieee-spectrum-robotics:96aa397c5848", "url": "https://spectrum.ieee.org/video-friday-reachy-mini-raps", "title": "Video Friday: Robot Decommissioning Takes a Fun Turn", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: IEEE Spectrum Robotics · 2026-10-09
It’s possible that I wasn’t the only one who thought that Figure’s recent stunt was somewhat ridiculous. Devanthro is in fact selling these plastic balls for $1000 each, and they’ll arrive in time for Christmas.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: IEEE Spectrum Robotics · 2026-10-09\n\nIt’s possible that I wasn’t the only one who thought that Figure’s recent stunt was somewhat ridiculous. Devanthro is in fact selling these plastic balls for $1000 each, and they’ll arrive in time for Christmas.\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "It’s possible that I wasn’t the only one who thought that Figure’s recent stunt was somewhat ridiculous. Devanthro is in fact selling these plastic balls for $1000 each, and they’ll arrive in time for Christmas.", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "IEEE Spectrum Robotics" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "ieee-spectrum-robotics:96aa397c5848", "url": "https://spectrum.ieee.org/video-friday-reachy-mini-raps", "title": "Video Friday: Robot Decommissioning Takes a Fun Turn", "publisher": "IEEE Spectrum Robotics", "source_id": "ieee-spectrum-robotics", "evidence": "M", "lane": "hardware", "published": "2026-10-09", "radar_published": "2026-10-10T07:15:00Z", "numbers": [], "signals": [], "excerpt": "It’s possible that I wasn’t the only one who thought that Figure’s recent stunt was somewhat ridiculous. Devanthro is in fact selling these plastic balls for $1000 each, and they’ll arrive in time for Christmas." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:4522d1bc988b", "url": "https://www.leiphone.com/category/robot/Pz34omTBnW9Y0LhO.html", "title": "让触觉像视觉一样可规模化，哥大李昀烛团队 FlexiTac 的开源实践 | IROS 2026", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-10
FlexiTac压阻式传感器可兼顾柔性、低成本与一致性。机器人的视觉能力正快速增强，但 触觉仍然是具身智能系统中的薄弱环节 。\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-10\n\nFlexiTac压阻式传感器可兼顾柔性、低成本与一致性。机器人的视觉能力正快速增强，但 触觉仍然是具身智能系统中的薄弱环节 。\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "FlexiTac压阻式传感器可兼顾柔性、低成本与一致性。机器人的视觉能力正快速增强，但 触觉仍然是具身智能系统中的薄弱环节 。", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "leiphone:4522d1bc988b", "url": "https://www.leiphone.com/category/robot/Pz34omTBnW9Y0LhO.html", "title": "让触觉像视觉一样可规模化，哥大李昀烛团队 FlexiTac 的开源实践 | IROS 2026", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "hardware", "published": "2026-10-10", "radar_published": "2026-10-10T07:15:00Z", "numbers": [], "signals": [], "excerpt": "FlexiTac压阻式传感器可兼顾柔性、低成本与一致性。机器人的视觉能力正快速增强，但 触觉仍然是具身智能系统中的薄弱环节 。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:2fc129443dc7", "url": "https://www.leiphone.com/category/ai/4I8YKksz9V10Ih1R.html", "title": "这届 IROS 2026 上，具身学术与产业边界正在消融", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-10
模型、数据、灵巧操作都在工程化，顶会上具身传递出的信号变了9 月 27 日到 10 月 1 日，全球机器人领域最顶级的国际会议之一 IROS 2026 将在美国匹兹堡开幕。 如果你前两年也在关注这个会，你会发现那时候大家都在 “卷概念”。\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-10\n\n模型、数据、灵巧操作都在工程化，顶会上具身传递出的信号变了9 月 27 日到 10 月 1 日，全球机器人领域最顶级的国际会议之一 IROS 2026 将在美国匹兹堡开幕。 如果你前两年也在关注这个会，你会发现那时候大家都在 “卷概念”。\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "模型、数据、灵巧操作都在工程化，顶会上具身传递出的信号变了9 月 27 日到 10 月 1 日，全球机器人领域最顶级的国际会议之一 IROS 2026 将在美国匹兹堡开幕。 如果你前两年也在关注这个会，你会发现那时候大家都在 “卷概念”。", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Data engine (collection / curation / scaling)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "leiphone:2fc129443dc7", "url": "https://www.leiphone.com/category/ai/4I8YKksz9V10Ih1R.html", "title": "这届 IROS 2026 上，具身学术与产业边界正在消融", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "data", "published": "2026-10-10", "radar_published": "2026-10-10T07:15:00Z", "numbers": [], "signals": [], "excerpt": "模型、数据、灵巧操作都在工程化，顶会上具身传递出的信号变了9 月 27 日到 10 月 1 日，全球机器人领域最顶级的国际会议之一 IROS 2026 将在美国匹兹堡开幕。 如果你前两年也在关注这个会，你会发现那时候大家都在 “卷概念”。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:6fcc505a20cb", "url": "https://www.leiphone.com/category/zaobao/GGmq1RiBgXY2GSVL.html", "title": "再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20%；腾讯拟发行50亿美元离岸债券，或筹钱加码AI；OpenAI全面上线GPT-6", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-09
要闻提示 再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20% 阿里巴巴蔡崇信：开源是欧洲发展 AI 的最佳选择 腾讯拟发行50亿美元离岸债券，或筹钱加码AI 超33亿元！\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-09\n\n要闻提示 再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20% 阿里巴巴蔡崇信：开源是欧洲发展 AI 的最佳选择 腾讯拟发行50亿美元离岸债券，或筹钱加码AI 超33亿元！\n\nFull radar for 2026-10-10: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-10.en.html", "summary": "要闻提示 再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20% 阿里巴巴蔡崇信：开源是欧洲发展 AI 的最佳选择 腾讯拟发行50亿美元离岸债券，或筹钱加码AI 超33亿元！", "date_published": "2026-10-10T07:15:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Foundation models (VLA / WAM)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-10", "id": "leiphone:6fcc505a20cb", "url": "https://www.leiphone.com/category/zaobao/GGmq1RiBgXY2GSVL.html", "title": "再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20%；腾讯拟发行50亿美元离岸债券，或筹钱加码AI；OpenAI全面上线GPT-6", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "foundation", "published": "2026-10-09", "radar_published": "2026-10-10T07:15:00Z", "numbers": [], "signals": [], "excerpt": "要闻提示 再裁员5000人！保时捷明确回归燃油车，高端车型计划涨价20% 阿里巴巴蔡崇信：开源是欧洲发展 AI 的最佳选择 腾讯拟发行50亿美元离岸债券，或筹钱加码AI 超33亿元！" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11508", "url": "https://arxiv.org/abs/2610.11508", "title": "WARP-VLA: Wrist-Camera Adaptation for View-Robust Policy Execution in Vision-Language-Action Models", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.11508 · 2026-10-08
Key numbers: 39.2% · 78.3%
Signals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights
Despite recent advances in Vision-Language-Action models (VLAs) for robotic manipulation, their performance remains sensitive to changes in camera configuration.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.11508 · 2026-10-08\nKey numbers: 39.2% · 78.3%\nSignals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights\n\nDespite recent advances in Vision-Language-Action models (VLAs) for robotic manipulation, their performance remains sensitive to changes in camera configuration.\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "Despite recent advances in Vision-Language-Action models (VLAs) for robotic manipulation, their performance remains sensitive to changes in camera configuration.", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "arXiv 2610.11508" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "arxiv:2610.11508", "url": "https://arxiv.org/abs/2610.11508", "title": "WARP-VLA: Wrist-Camera Adaptation for View-Robust Policy Execution in Vision-Language-Action Models", "publisher": "arXiv 2610.11508", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [ "39.2%", "78.3%" ], "signals": [ "numbers", "closed_loop", "real_robot", "open_release" ], "excerpt": "Despite recent advances in Vision-Language-Action models (VLAs) for robotic manipulation, their performance remains sensitive to changes in camera configuration." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12007", "url": "https://arxiv.org/abs/2610.12007", "title": "REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.12007 · 2026-10-08
Signals: closed-loop success reported · real-robot experiments · latency or control-rate numbers
Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.12007 · 2026-10-08\nSignals: closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nFlow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning…\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning…", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "arXiv 2610.12007" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "arxiv:2610.12007", "url": "https://arxiv.org/abs/2610.12007", "title": "REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models", "publisher": "arXiv 2610.12007", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot", "latency" ], "excerpt": "Flow-based vision-language-action (VLA) models generate action chunks for temporally coherent robot motion, but chunked control creates a fundamental closed-loop trade-off: long chunks provide smooth execution, whereas frequent replanning…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.12140", "url": "https://arxiv.org/abs/2610.12140", "title": "Leveraging Human-In-The-Loop Demonstrations in Reinforcement Learning for Digital Twin-Driven Robot Flexibility", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [R]
Source: arXiv 2610.12140 · 2026-10-08
Key numbers: 100% · 17%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Growing automation makes collaborative robots work in more variable environments, increasing the need for adaptation.\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [R]\nSource: arXiv 2610.12140 · 2026-10-08\nKey numbers: 100% · 17%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nGrowing automation makes collaborative robots work in more variable environments, increasing the need for adaptation.\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "Growing automation makes collaborative robots work in more variable environments, increasing the need for adaptation.", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "arXiv 2610.12140" } ], "tags": [ "Data engine (collection / curation / scaling)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "arxiv:2610.12140", "url": "https://arxiv.org/abs/2610.12140", "title": "Leveraging Human-In-The-Loop Demonstrations in Reinforcement Learning for Digital Twin-Driven Robot Flexibility", "publisher": "arXiv 2610.12140", "source_id": "arxiv", "evidence": "R", "lane": "data", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [ "100%", "17%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Growing automation makes collaborative robots work in more variable environments, increasing the need for adaptation." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.11952", "url": "https://arxiv.org/abs/2610.11952", "title": "Tell Robot What Not to Do: A Negation Understanding Perspective", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.11952 · 2026-10-08
Key numbers: 2.60% · 88.45% · 12.4% · 88.8%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Instruction following enables robots to perform diverse tasks specified in natural language, making it a fundamental capability for human-robot interaction.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.11952 · 2026-10-08\nKey numbers: 2.60% · 88.45% · 12.4% · 88.8%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nInstruction following enables robots to perform diverse tasks specified in natural language, making it a fundamental capability for human-robot interaction.\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "Instruction following enables robots to perform diverse tasks specified in natural language, making it a fundamental capability for human-robot interaction.", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "arXiv 2610.11952" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "arxiv:2610.11952", "url": "https://arxiv.org/abs/2610.11952", "title": "Tell Robot What Not to Do: A Negation Understanding Perspective", "publisher": "arXiv 2610.11952", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [ "2.60%", "88.45%", "12.4%", "88.8%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Instruction following enables robots to perform diverse tasks specified in natural language, making it a fundamental capability for human-robot interaction." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:ieee-spectrum-robotics:a2b57285998e", "url": "https://spectrum.ieee.org/walking-robotic-hand", "title": "This Disembodied Hand Is All the Robot You Need", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: IEEE Spectrum Robotics · 2026-10-08
A hand attached to the end of an arm is a given for most humans, but what if robots didn’t have to play by the same rules as we do?\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: IEEE Spectrum Robotics · 2026-10-08\n\nA hand attached to the end of an arm is a given for most humans, but what if robots didn’t have to play by the same rules as we do?\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "A hand attached to the end of an arm is a given for most humans, but what if robots didn’t have to play by the same rules as we do?", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "IEEE Spectrum Robotics" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "ieee-spectrum-robotics:a2b57285998e", "url": "https://spectrum.ieee.org/walking-robotic-hand", "title": "This Disembodied Hand Is All the Robot You Need", "publisher": "IEEE Spectrum Robotics", "source_id": "ieee-spectrum-robotics", "evidence": "M", "lane": "hardware", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [], "signals": [], "excerpt": "A hand attached to the end of an arm is a given for most humans, but what if robots didn’t have to play by the same rules as we do?" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:bcb1ebbc55ff", "url": "https://www.therobotreport.com/jabil-discusses-pace-humanoid-robot-development-production/", "title": "Jabil discusses the pace of humanoid robot development and production", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: The Robot Report · 2026-10-08
Jabil, which is working with robot developers such as Apptronik, provides a design perspective on humanoid progress.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: The Robot Report · 2026-10-08\n\nJabil, which is working with robot developers such as Apptronik, provides a design perspective on humanoid progress.\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "Jabil, which is working with robot developers such as Apptronik, provides a design perspective on humanoid progress.", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "the-robot-report:bcb1ebbc55ff", "url": "https://www.therobotreport.com/jabil-discusses-pace-humanoid-robot-development-production/", "title": "Jabil discusses the pace of humanoid robot development and production", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "hardware", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [], "signals": [], "excerpt": "Jabil, which is working with robot developers such as Apptronik, provides a design perspective on humanoid progress." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:6b2f68612d16", "url": "https://www.therobotreport.com/aws-launches-open-source-physical-ai-toolchain-for-robotics/", "title": "AWS launches open-source Physical AI Toolchain for robotics", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [M]
Source: The Robot Report · 2026-10-08
Signals: open code or weights
The toolchain combines AWS and NVIDIA tools for synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [M]\nSource: The Robot Report · 2026-10-08\nSignals: open code or weights\n\nThe toolchain combines AWS and NVIDIA tools for synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement.\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "The toolchain combines AWS and NVIDIA tools for synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement.", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "the-robot-report:6b2f68612d16", "url": "https://www.therobotreport.com/aws-launches-open-source-physical-ai-toolchain-for-robotics/", "title": "AWS launches open-source Physical AI Toolchain for robotics", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "edge", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [], "signals": [ "open_release" ], "excerpt": "The toolchain combines AWS and NVIDIA tools for synthetic data generation, model training, simulation and validation, edge deployment, and continuous improvement." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:df0a0c0416fd", "url": "https://www.leiphone.com/category/ai/WbD0Tf6P1oA2WDdq.html", "title": "独家 | 清华 AIR 首届博士李健雄创立本溯智能：五源资本领投，押注动作原生具身模型", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-08
常春藤、线性、华山资本跟投，AIR副教授詹仙园任首席科学家。雷峰网·鲸犀独家获悉，清华大学智能产业研究院（AIR）首届博士生李健雄已毕业，并创立具身智能公司 本溯智能 （BASAL Intelligence，简称 BASAL）。\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-08\n\n常春藤、线性、华山资本跟投，AIR副教授詹仙园任首席科学家。雷峰网·鲸犀独家获悉，清华大学智能产业研究院（AIR）首届博士生李健雄已毕业，并创立具身智能公司 本溯智能 （BASAL Intelligence，简称 BASAL）。\n\nFull radar for 2026-10-09: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-09.en.html", "summary": "常春藤、线性、华山资本跟投，AIR副教授詹仙园任首席科学家。雷峰网·鲸犀独家获悉，清华大学智能产业研究院（AIR）首届博士生李健雄已毕业，并创立具身智能公司 本溯智能 （BASAL Intelligence，简称 BASAL）。", "date_published": "2026-10-09T07:28:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-09", "id": "leiphone:df0a0c0416fd", "url": "https://www.leiphone.com/category/ai/WbD0Tf6P1oA2WDdq.html", "title": "独家 | 清华 AIR 首届博士李健雄创立本溯智能：五源资本领投，押注动作原生具身模型", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "training", "published": "2026-10-08", "radar_published": "2026-10-09T07:28:00Z", "numbers": [], "signals": [], "excerpt": "常春藤、线性、华山资本跟投，AIR副教授詹仙园任首席科学家。雷峰网·鲸犀独家获悉，清华大学智能产业研究院（AIR）首届博士生李健雄已毕业，并创立具身智能公司 本溯智能 （BASAL Intelligence，简称 BASAL）。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.10479", "url": "https://arxiv.org/abs/2610.10479", "title": "Agentic RSR: Real-to-Sim-to-Real through Scene Reconstruction and Execution-Grounded Robot Policies", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.10479 · 2026-10-07
Key numbers: 80%
Signals: quantitative results · closed-loop success reported · real-robot experiments
A simulation of a real robot workspace must preserve task-relevant interactions, while policies developed in it must operate on observations available to the real robot.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.10479 · 2026-10-07\nKey numbers: 80%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nA simulation of a real robot workspace must preserve task-relevant interactions, while policies developed in it must operate on observations available to the real robot.\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "A simulation of a real robot workspace must preserve task-relevant interactions, while policies developed in it must operate on observations available to the real robot.", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "arXiv 2610.10479" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "arxiv:2610.10479", "url": "https://arxiv.org/abs/2610.10479", "title": "Agentic RSR: Real-to-Sim-to-Real through Scene Reconstruction and Execution-Grounded Robot Policies", "publisher": "arXiv 2610.10479", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-07", "radar_published": "2026-10-08T07:30:00Z", "numbers": [ "80%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "A simulation of a real robot workspace must preserve task-relevant interactions, while policies developed in it must operate on observations available to the real robot." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.10079", "url": "https://arxiv.org/abs/2610.10079", "title": "RealtimeWAM: How Fast Can I Run My World Action Model?", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.10079 · 2026-10-07
Key numbers: 63.09 ms · 82.75% · 87.41%
Signals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers
World Action Models (WAMs) combine visual dynamics modeling with action generation, but their high inference latency limits responsive robot control.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.10079 · 2026-10-07\nKey numbers: 63.09 ms · 82.75% · 87.41%\nSignals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nWorld Action Models (WAMs) combine visual dynamics modeling with action generation, but their high inference latency limits responsive robot control.\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "World Action Models (WAMs) combine visual dynamics modeling with action generation, but their high inference latency limits responsive robot control.", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "arXiv 2610.10079" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "arxiv:2610.10079", "url": "https://arxiv.org/abs/2610.10079", "title": "RealtimeWAM: How Fast Can I Run My World Action Model?", "publisher": "arXiv 2610.10079", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-07", "radar_published": "2026-10-08T07:30:00Z", "numbers": [ "63.09 ms", "82.75%", "87.41%" ], "signals": [ "numbers", "closed_loop", "real_robot", "latency" ], "excerpt": "World Action Models (WAMs) combine visual dynamics modeling with action generation, but their high inference latency limits responsive robot control." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.10453", "url": "https://arxiv.org/abs/2610.10453", "title": "RFPO: Rectified Flow Policy Optimization for Embodied Control", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.10453 · 2026-10-07
Key numbers: 2.4% · 98.5% · 4.39 ms · 0.08 ms
Signals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers
Flow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.10453 · 2026-10-07\nKey numbers: 2.4% · 98.5% · 4.39 ms · 0.08 ms\nSignals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers\n\nFlow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost.\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "Flow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost.", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "arXiv 2610.10453" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "arxiv:2610.10453", "url": "https://arxiv.org/abs/2610.10453", "title": "RFPO: Rectified Flow Policy Optimization for Embodied Control", "publisher": "arXiv 2610.10453", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-07", "radar_published": "2026-10-08T07:30:00Z", "numbers": [ "2.4%", "98.5%", "4.39 ms", "0.08 ms" ], "signals": [ "numbers", "real_robot", "open_release", "latency" ], "excerpt": "Flow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.08784", "url": "https://arxiv.org/abs/2610.08784", "title": "PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.08784 · 2026-10-06
Key numbers: 53.2% · 95% · 90%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Pretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.08784 · 2026-10-06\nKey numbers: 53.2% · 95% · 90%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nPretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction.\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "Pretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction.", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "arXiv 2610.08784" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "arxiv:2610.08784", "url": "https://arxiv.org/abs/2610.08784", "title": "PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation", "publisher": "arXiv 2610.08784", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-10-06", "radar_published": "2026-10-08T07:30:00Z", "numbers": [ "53.2%", "95%", "90%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Pretrained robotic policies can suffer substantial performance degradation under out-of-distribution (OOD) conditions encountered during deployment, motivating post-training through real-world interaction." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:4b07be24d804", "url": "https://www.leiphone.com/category/private/wEpCJ4T8v9cJFgCn.html", "title": "研究了1933篇 IROS 论文，我们看到了机器人学的六项新变化", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-08
大模型并没有“吃掉”机器人学。9月27日，IROS 2026 将在匹兹堡开幕。\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-08\n\n大模型并没有“吃掉”机器人学。9月27日，IROS 2026 将在匹兹堡开幕。\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "大模型并没有“吃掉”机器人学。9月27日，IROS 2026 将在匹兹堡开幕。", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "leiphone:4b07be24d804", "url": "https://www.leiphone.com/category/private/wEpCJ4T8v9cJFgCn.html", "title": "研究了1933篇 IROS 论文，我们看到了机器人学的六项新变化", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "hardware", "published": "2026-10-08", "radar_published": "2026-10-08T07:30:00Z", "numbers": [], "signals": [], "excerpt": "大模型并没有“吃掉”机器人学。9月27日，IROS 2026 将在匹兹堡开幕。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:f05eb6e1da2e", "url": "https://www.leiphone.com/category/transportation/ESCXNijR4DVx6uDl.html", "title": "对话如祺出行COO韩锋：世界模型越强，真实数据才是真壁垒", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-08
过去一年，“物理AI”几乎成了自动驾驶、车企共同争夺的下一张船票。 世界模型、具身大模型、机器人数据……众多公司从不同切入口布局物理AI，试图成为新风口之下的主角，相比之下，如祺出行选择了一条不那么显眼的路径，不自己训练大模型，而是尝试成为模型背后的数据基础设施。\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-08\n\n过去一年，“物理AI”几乎成了自动驾驶、车企共同争夺的下一张船票。 世界模型、具身大模型、机器人数据……众多公司从不同切入口布局物理AI，试图成为新风口之下的主角，相比之下，如祺出行选择了一条不那么显眼的路径，不自己训练大模型，而是尝试成为模型背后的数据基础设施。\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "过去一年，“物理AI”几乎成了自动驾驶、车企共同争夺的下一张船票。 世界模型、具身大模型、机器人数据……众多公司从不同切入口布局物理AI，试图成为新风口之下的主角，相比之下，如祺出行选择了一条不那么显眼的路径，不自己训练大模型，而是尝试成为模型背后的数据基础设施。", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Data engine (collection / curation / scaling)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "leiphone:f05eb6e1da2e", "url": "https://www.leiphone.com/category/transportation/ESCXNijR4DVx6uDl.html", "title": "对话如祺出行COO韩锋：世界模型越强，真实数据才是真壁垒", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "data", "published": "2026-10-08", "radar_published": "2026-10-08T07:30:00Z", "numbers": [], "signals": [], "excerpt": "过去一年，“物理AI”几乎成了自动驾驶、车企共同争夺的下一张船票。 世界模型、具身大模型、机器人数据……众多公司从不同切入口布局物理AI，试图成为新风口之下的主角，相比之下，如祺出行选择了一条不那么显眼的路径，不自己训练大模型，而是尝试成为模型背后的数据基础设施。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:leiphone:bee8b4023118", "url": "https://www.leiphone.com/category/private/VYRZFCnadDOcPDeX.html", "title": "IROS 2026大奖揭晓：长期记忆拿下最佳论文，人形机器人打网球、端托盘双双获奖", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [M]
Source: 雷峰网 Leiphone · 2026-10-08
在机器人领域，一个更大的模型并不能自动解决所有问题。当地时间 9 月 30 日，IROS 2026 主会在美国匹兹堡落下帷幕，本届大会的主要奖项也在最后一天正式揭晓。\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [M]\nSource: 雷峰网 Leiphone · 2026-10-08\n\n在机器人领域，一个更大的模型并不能自动解决所有问题。当地时间 9 月 30 日，IROS 2026 主会在美国匹兹堡落下帷幕，本届大会的主要奖项也在最后一天正式揭晓。\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "在机器人领域，一个更大的模型并不能自动解决所有问题。当地时间 9 月 30 日，IROS 2026 主会在美国匹兹堡落下帷幕，本届大会的主要奖项也在最后一天正式揭晓。", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "雷峰网 Leiphone" } ], "tags": [ "Foundation models (VLA / WAM)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "leiphone:bee8b4023118", "url": "https://www.leiphone.com/category/private/VYRZFCnadDOcPDeX.html", "title": "IROS 2026大奖揭晓：长期记忆拿下最佳论文，人形机器人打网球、端托盘双双获奖", "publisher": "雷峰网 Leiphone", "source_id": "leiphone", "evidence": "M", "lane": "foundation", "published": "2026-10-08", "radar_published": "2026-10-08T07:30:00Z", "numbers": [], "signals": [], "excerpt": "在机器人领域，一个更大的模型并不能自动解决所有问题。当地时间 9 月 30 日，IROS 2026 主会在美国匹兹堡落下帷幕，本届大会的主要奖项也在最后一天正式揭晓。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:42327a98fe3f", "url": "https://www.therobotreport.com/harmonic-drive-share-how-integrated-actuators-reduce-engineering-robobusiness/", "title": "Harmonic Drive to share how integrated actuators can reduce engineering at RoboBusiness", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: The Robot Report · 2026-10-07
Attendees will learn how integrated actuators can reduce engineering effort by simplifying system integration.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: The Robot Report · 2026-10-07\n\nAttendees will learn how integrated actuators can reduce engineering effort by simplifying system integration.\n\nFull radar for 2026-10-08: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-08.en.html", "summary": "Attendees will learn how integrated actuators can reduce engineering effort by simplifying system integration.", "date_published": "2026-10-08T07:30:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-08", "id": "the-robot-report:42327a98fe3f", "url": "https://www.therobotreport.com/harmonic-drive-share-how-integrated-actuators-reduce-engineering-robobusiness/", "title": "Harmonic Drive to share how integrated actuators can reduce engineering at RoboBusiness", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "hardware", "published": "2026-10-07", "radar_published": "2026-10-08T07:30:00Z", "numbers": [], "signals": [], "excerpt": "Attendees will learn how integrated actuators can reduce engineering effort by simplifying system integration." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.08618", "url": "https://arxiv.org/abs/2610.08618", "title": "Demo: Closed-Loop Sionna-Isaac Sim Co-Simulation Framework for Wireless-Aware Robot Navigation over ROS 2", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.08618 · 2026-10-06
Key numbers: 16ms · 60 Hz · 7.4%
Signals: quantitative results · closed-loop success reported · latency or control-rate numbers
A robot that offloads its control loop to the network carries the receiver with it, so link quality is decided by where it goes.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.08618 · 2026-10-06\nKey numbers: 16ms · 60 Hz · 7.4%\nSignals: quantitative results · closed-loop success reported · latency or control-rate numbers\n\nA robot that offloads its control loop to the network carries the receiver with it, so link quality is decided by where it goes.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "A robot that offloads its control loop to the network carries the receiver with it, so link quality is decided by where it goes.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.08618" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.08618", "url": "https://arxiv.org/abs/2610.08618", "title": "Demo: Closed-Loop Sionna-Isaac Sim Co-Simulation Framework for Wireless-Aware Robot Navigation over ROS 2", "publisher": "arXiv 2610.08618", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [ "16ms", "60 Hz", "7.4%" ], "signals": [ "numbers", "closed_loop", "latency" ], "excerpt": "A robot that offloads its control loop to the network carries the receiver with it, so link quality is decided by where it goes." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.07756", "url": "https://arxiv.org/abs/2610.07756", "title": "StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.07756 · 2026-10-06
Key numbers: 96.5% · 97.8% · 115.0 ms · 44.2 ms
Signals: quantitative results · closed-loop success reported · latency or control-rate numbers
Vision-language-action (VLA) models increasingly rely on diffusion- or flow-matching-based action heads to generate continuous robot actions.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.07756 · 2026-10-06\nKey numbers: 96.5% · 97.8% · 115.0 ms · 44.2 ms\nSignals: quantitative results · closed-loop success reported · latency or control-rate numbers\n\nVision-language-action (VLA) models increasingly rely on diffusion- or flow-matching-based action heads to generate continuous robot actions.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Vision-language-action (VLA) models increasingly rely on diffusion- or flow-matching-based action heads to generate continuous robot actions.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.07756" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.07756", "url": "https://arxiv.org/abs/2610.07756", "title": "StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models", "publisher": "arXiv 2610.07756", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [ "96.5%", "97.8%", "115.0 ms", "44.2 ms" ], "signals": [ "numbers", "closed_loop", "latency" ], "excerpt": "Vision-language-action (VLA) models increasingly rely on diffusion- or flow-matching-based action heads to generate continuous robot actions." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.07696", "url": "https://arxiv.org/abs/2610.07696", "title": "ESP: Energy-Score Policy for One-Step Multimodal Action Generation", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.07696 · 2026-10-06
Signals: closed-loop success reported · real-robot experiments · latency or control-rate numbers
Generative action models based on diffusion and flow matching have been increasingly adopted in vision-language-action (VLA) policies for their ability to capture diverse behaviors, including multiple valid action sequences under the same…\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.07696 · 2026-10-06\nSignals: closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nGenerative action models based on diffusion and flow matching have been increasingly adopted in vision-language-action (VLA) policies for their ability to capture diverse behaviors, including multiple valid action sequences under the same…\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Generative action models based on diffusion and flow matching have been increasingly adopted in vision-language-action (VLA) policies for their ability to capture diverse behaviors, including multiple valid action sequences under the same…", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.07696" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.07696", "url": "https://arxiv.org/abs/2610.07696", "title": "ESP: Energy-Score Policy for One-Step Multimodal Action Generation", "publisher": "arXiv 2610.07696", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot", "latency" ], "excerpt": "Generative action models based on diffusion and flow matching have been increasingly adopted in vision-language-action (VLA) policies for their ability to capture diverse behaviors, including multiple valid action sequences under the same…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.07946", "url": "https://arxiv.org/abs/2610.07946", "title": "Adapting Vision-Language-Action Models to Unknown Visual Disruptions During Execution", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.07946 · 2026-10-06
Key numbers: 43.9% · 53.2% · 58.7% · 66.0%
Signals: quantitative results · real-robot experiments
Visual disruptions can arise while a robot is executing a task, leaving a vision-language-action (VLA) policy to respond without knowing the disruption type or timing.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.07946 · 2026-10-06\nKey numbers: 43.9% · 53.2% · 58.7% · 66.0%\nSignals: quantitative results · real-robot experiments\n\nVisual disruptions can arise while a robot is executing a task, leaving a vision-language-action (VLA) policy to respond without knowing the disruption type or timing.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Visual disruptions can arise while a robot is executing a task, leaving a vision-language-action (VLA) policy to respond without knowing the disruption type or timing.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.07946" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.07946", "url": "https://arxiv.org/abs/2610.07946", "title": "Adapting Vision-Language-Action Models to Unknown Visual Disruptions During Execution", "publisher": "arXiv 2610.07946", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [ "43.9%", "53.2%", "58.7%", "66.0%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Visual disruptions can arise while a robot is executing a task, leaving a vision-language-action (VLA) policy to respond without knowing the disruption type or timing." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.05024", "url": "https://arxiv.org/abs/2610.05024", "title": "LightVLN: Efficient Aerial Vision-and-Language Navigation with Compact Memory and History-Guided Local Aggregation", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.05024 · 2026-10-04
Key numbers: 50.93% · 36.14% · 25.83% · 16 GB
Signals: quantitative results · closed-loop success reported · latency or control-rate numbers
Aerial vision-and-language navigation (VLN) enables unmanned aerial vehicles to execute long-horizon natural-language instructions from visual observations in complex three-dimensional environments.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.05024 · 2026-10-04\nKey numbers: 50.93% · 36.14% · 25.83% · 16 GB\nSignals: quantitative results · closed-loop success reported · latency or control-rate numbers\n\nAerial vision-and-language navigation (VLN) enables unmanned aerial vehicles to execute long-horizon natural-language instructions from visual observations in complex three-dimensional environments.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Aerial vision-and-language navigation (VLN) enables unmanned aerial vehicles to execute long-horizon natural-language instructions from visual observations in complex three-dimensional environments.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.05024" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.05024", "url": "https://arxiv.org/abs/2610.05024", "title": "LightVLN: Efficient Aerial Vision-and-Language Navigation with Compact Memory and History-Guided Local Aggregation", "publisher": "arXiv 2610.05024", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-04", "radar_published": "2026-10-07T07:19:00Z", "numbers": [ "50.93%", "36.14%", "25.83%", "16 GB" ], "signals": [ "numbers", "closed_loop", "latency" ], "excerpt": "Aerial vision-and-language navigation (VLN) enables unmanned aerial vehicles to execute long-horizon natural-language instructions from visual observations in complex three-dimensional environments." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.08444", "url": "https://arxiv.org/abs/2610.08444", "title": "ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.08444 · 2026-10-06
Signals: closed-loop success reported · latency or control-rate numbers
Vision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.08444 · 2026-10-06\nSignals: closed-loop success reported · latency or control-rate numbers\n\nVision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Vision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "arXiv 2610.08444" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "arxiv:2610.08444", "url": "https://arxiv.org/abs/2610.08444", "title": "ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference", "publisher": "arXiv 2610.08444", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [], "signals": [ "closed_loop", "latency" ], "excerpt": "Vision-language-action (VLA) policies repeatedly invoke inference to control robots, making graphics processing unit (GPU) energy a recurring cost of task execution." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:monoist:41b504f26025", "url": "https://monoist.itmedia.co.jp/mn/articles/2610/07/news043.html", "title": "ルネサスが低耐圧GaNデバイス市場に参入、業界トップクラスの性能指数を実現", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: MONOist (ITmedia) · 2026-10-07
ルネサス エレクトロニクスは、同社初となる100V耐圧のEモードGaNパワートランジスタ製品群を発売した。AIデータセンターやヒューマノイドロボットなど向けに、超高速スイッチングと優れた放熱性能を実現する。\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: MONOist (ITmedia) · 2026-10-07\n\nルネサス エレクトロニクスは、同社初となる100V耐圧のEモードGaNパワートランジスタ製品群を発売した。AIデータセンターやヒューマノイドロボットなど向けに、超高速スイッチングと優れた放熱性能を実現する。\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "ルネサス エレクトロニクスは、同社初となる100V耐圧のEモードGaNパワートランジスタ製品群を発売した。AIデータセンターやヒューマノイドロボットなど向けに、超高速スイッチングと優れた放熱性能を実現する。", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "MONOist (ITmedia)" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "monoist:41b504f26025", "url": "https://monoist.itmedia.co.jp/mn/articles/2610/07/news043.html", "title": "ルネサスが低耐圧GaNデバイス市場に参入、業界トップクラスの性能指数を実現", "publisher": "MONOist (ITmedia)", "source_id": "monoist", "evidence": "M", "lane": "hardware", "published": "2026-10-07", "radar_published": "2026-10-07T07:19:00Z", "numbers": [], "signals": [], "excerpt": "ルネサス エレクトロニクスは、同社初となる100V耐圧のEモードGaNパワートランジスタ製品群を発売した。AIデータセンターやヒューマノイドロボットなど向けに、超高速スイッチングと優れた放熱性能を実現する。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:04735ae4ac05", "url": "https://www.therobotreport.com/boston-dynamics-appoints-former-amazon-executive-rohit-prasad-new-ceo/", "title": "Boston Dynamics appoints former Amazon exec Rohit Prasad as new CEO", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: The Robot Report · 2026-10-06
Before Boston Dynamics, Prasad served as senior vice president and head scientist for Alexa and artificial general intelligence at Amazon.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: The Robot Report · 2026-10-06\n\nBefore Boston Dynamics, Prasad served as senior vice president and head scientist for Alexa and artificial general intelligence at Amazon.\n\nFull radar for 2026-10-07: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-07.en.html", "summary": "Before Boston Dynamics, Prasad served as senior vice president and head scientist for Alexa and artificial general intelligence at Amazon.", "date_published": "2026-10-07T07:19:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-07", "id": "the-robot-report:04735ae4ac05", "url": "https://www.therobotreport.com/boston-dynamics-appoints-former-amazon-executive-rohit-prasad-new-ceo/", "title": "Boston Dynamics appoints former Amazon exec Rohit Prasad as new CEO", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "hardware", "published": "2026-10-06", "radar_published": "2026-10-07T07:19:00Z", "numbers": [], "signals": [], "excerpt": "Before Boston Dynamics, Prasad served as senior vice president and head scientist for Alexa and artificial general intelligence at Amazon." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.02832", "url": "https://arxiv.org/abs/2610.02832", "title": "FastOPD: On-Policy Distillation for Lightweight VLA Deployment", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.02832 · 2026-10-02
Key numbers: 84% · 78.1%
Signals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers
Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.02832 · 2026-10-02\nKey numbers: 84% · 78.1%\nSignals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nVision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.02832" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.02832", "url": "https://arxiv.org/abs/2610.02832", "title": "FastOPD: On-Policy Distillation for Lightweight VLA Deployment", "publisher": "arXiv 2610.02832", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "84%", "78.1%" ], "signals": [ "numbers", "closed_loop", "real_robot", "latency" ], "excerpt": "Vision-Language-Action (VLA) foundation models have scaled rapidly to enhance manipulation performance and generalizability, but this scaling incurs high computational costs that render real-world deployment increasingly challenging." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.03620", "url": "https://arxiv.org/abs/2610.03620", "title": "UniIntervene++: An Adaptive Intervention Agent for Efficient Real-World Reinforcement Learning", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.03620 · 2026-10-02
Key numbers: 89.67% · 0.77% · 94.6%
Signals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights
Online reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.03620 · 2026-10-02\nKey numbers: 89.67% · 0.77% · 94.6%\nSignals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights\n\nOnline reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Online reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.03620" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.03620", "url": "https://arxiv.org/abs/2610.03620", "title": "UniIntervene++: An Adaptive Intervention Agent for Efficient Real-World Reinforcement Learning", "publisher": "arXiv 2610.03620", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "89.67%", "0.77%", "94.6%" ], "signals": [ "numbers", "closed_loop", "real_robot", "open_release" ], "excerpt": "Online reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.03476", "url": "https://arxiv.org/abs/2610.03476", "title": "MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation", "content_html": "
Lane: Systems & orchestration (long-horizon / multi-policy)
Evidence: [R]
Source: arXiv 2610.03476 · 2026-10-02
Key numbers: 7.50% · 27.50% · 32.5% · 57.5%
Signals: quantitative results · real-robot experiments
Long-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control.\n", "content_text": "Lane: Systems & orchestration (long-horizon / multi-policy)\nEvidence: [R]\nSource: arXiv 2610.03476 · 2026-10-02\nKey numbers: 7.50% · 27.50% · 32.5% · 57.5%\nSignals: quantitative results · real-robot experiments\n\nLong-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Long-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.03476" } ], "tags": [ "Systems & orchestration (long-horizon / multi-policy)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.03476", "url": "https://arxiv.org/abs/2610.03476", "title": "MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation", "publisher": "arXiv 2610.03476", "source_id": "arxiv", "evidence": "R", "lane": "systems", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "7.50%", "27.50%", "32.5%", "57.5%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Long-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.03607", "url": "https://arxiv.org/abs/2610.03607", "title": "World Action Learning via Interaction-Centric Spectral Latent Guidance", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.03607 · 2026-10-02
Key numbers: 99.20% · 93.80% · 57.7%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Learning general-purpose robot policies requires large-scale real-world interaction data, yet collecting robot demonstrations remains expensive and difficult to scale.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.03607 · 2026-10-02\nKey numbers: 99.20% · 93.80% · 57.7%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nLearning general-purpose robot policies requires large-scale real-world interaction data, yet collecting robot demonstrations remains expensive and difficult to scale.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Learning general-purpose robot policies requires large-scale real-world interaction data, yet collecting robot demonstrations remains expensive and difficult to scale.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.03607" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.03607", "url": "https://arxiv.org/abs/2610.03607", "title": "World Action Learning via Interaction-Centric Spectral Latent Guidance", "publisher": "arXiv 2610.03607", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "99.20%", "93.80%", "57.7%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Learning general-purpose robot policies requires large-scale real-world interaction data, yet collecting robot demonstrations remains expensive and difficult to scale." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.02784", "url": "https://arxiv.org/abs/2610.02784", "title": "SimpleTouch: Can Vision-Language-Action Models Master Contact-Rich Manipulation Without Tactile Policy Pretraining?", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.02784 · 2026-10-02
Key numbers: 50 demonstrations · 77.5% · 45.2% · 66.7%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Tactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.02784 · 2026-10-02\nKey numbers: 50 demonstrations · 77.5% · 45.2% · 66.7%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nTactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Tactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.02784" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.02784", "url": "https://arxiv.org/abs/2610.02784", "title": "SimpleTouch: Can Vision-Language-Action Models Master Contact-Rich Manipulation Without Tactile Policy Pretraining?", "publisher": "arXiv 2610.02784", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "50 demonstrations", "77.5%", "45.2%", "66.7%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Tactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.03283", "url": "https://arxiv.org/abs/2610.03283", "title": "HexVIO: Towards All-Day Stereo-Inertial Tracking Through Commodity DSPs", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.03283 · 2026-10-02
Key numbers: 67% · 86% · 30 fps · 0.83 W
Signals: quantitative results · latency or control-rate numbers
The ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.03283 · 2026-10-02\nKey numbers: 67% · 86% · 30 fps · 0.83 W\nSignals: quantitative results · latency or control-rate numbers\n\nThe ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing.\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "The ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing.", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "arXiv 2610.03283" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "arxiv:2610.03283", "url": "https://arxiv.org/abs/2610.03283", "title": "HexVIO: Towards All-Day Stereo-Inertial Tracking Through Commodity DSPs", "publisher": "arXiv 2610.03283", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-02", "radar_published": "2026-10-05T23:40:00Z", "numbers": [ "67%", "86%", "30 fps", "0.83 W" ], "signals": [ "numbers", "latency" ], "excerpt": "The ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:077068b67d2b", "url": "https://www.therobotreport.com/fcc-robot-restrictions-could-accelerate-shift-to-local-ai/", "title": "FCC robot restrictions could accelerate shift to local AI", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [M]
Source: The Robot Report · 2026-10-05
Companies deploying physical AI today need to decide which AI workloads belong on the robot, what can remain in the cloud, and whether the underlying technology stack can meet rising expectations for security, safety, and supply-chain…\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [M]\nSource: The Robot Report · 2026-10-05\n\nCompanies deploying physical AI today need to decide which AI workloads belong on the robot, what can remain in the cloud, and whether the underlying technology stack can meet rising expectations for security, safety, and supply-chain…\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "Companies deploying physical AI today need to decide which AI workloads belong on the robot, what can remain in the cloud, and whether the underlying technology stack can meet rising expectations for security, safety, and supply-chain…", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Safety, permissions & compliance", "[M]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "the-robot-report:077068b67d2b", "url": "https://www.therobotreport.com/fcc-robot-restrictions-could-accelerate-shift-to-local-ai/", "title": "FCC robot restrictions could accelerate shift to local AI", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "safety", "published": "2026-10-05", "radar_published": "2026-10-05T23:40:00Z", "numbers": [], "signals": [], "excerpt": "Companies deploying physical AI today need to decide which AI workloads belong on the robot, what can remain in the cloud, and whether the underlying technology stack can meet rising expectations for security, safety, and supply-chain…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:monoist:7aaa16ae267f", "url": "https://monoist.itmedia.co.jp/mn/articles/2610/06/news009.html", "title": "川崎重工とNVIDIAが目指す「次世代デジタルシップヤード構想」とは何か", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [M]
Source: MONOist (ITmedia) · 2026-10-05
川崎重工は、NVIDIAのフィジカルAIとデジタルツイン技術を活用し、「次世代デジタルシップヤード」の実現を目指すプロジェクトを開始した。海外で進むデジタルシップヤード構想と比べて、川崎重工とNVIDIAの取り組みにはどのような違いがあるのだろうのか。\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [M]\nSource: MONOist (ITmedia) · 2026-10-05\n\n川崎重工は、NVIDIAのフィジカルAIとデジタルツイン技術を活用し、「次世代デジタルシップヤード」の実現を目指すプロジェクトを開始した。海外で進むデジタルシップヤード構想と比べて、川崎重工とNVIDIAの取り組みにはどのような違いがあるのだろうのか。\n\nFull radar for 2026-10-06: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-06.en.html", "summary": "川崎重工は、NVIDIAのフィジカルAIとデジタルツイン技術を活用し、「次世代デジタルシップヤード」の実現を目指すプロジェクトを開始した。海外で進むデジタルシップヤード構想と比べて、川崎重工とNVIDIAの取り組みにはどのような違いがあるのだろうのか。", "date_published": "2026-10-05T23:40:00Z", "authors": [ { "name": "MONOist (ITmedia)" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-06", "id": "monoist:7aaa16ae267f", "url": "https://monoist.itmedia.co.jp/mn/articles/2610/06/news009.html", "title": "川崎重工とNVIDIAが目指す「次世代デジタルシップヤード構想」とは何か", "publisher": "MONOist (ITmedia)", "source_id": "monoist", "evidence": "M", "lane": "simeval", "published": "2026-10-05", "radar_published": "2026-10-05T23:40:00Z", "numbers": [], "signals": [], "excerpt": "川崎重工は、NVIDIAのフィジカルAIとデジタルツイン技術を活用し、「次世代デジタルシップヤード」の実現を目指すプロジェクトを開始した。海外で進むデジタルシップヤード構想と比べて、川崎重工とNVIDIAの取り組みにはどのような違いがあるのだろうのか。" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.01351", "url": "https://arxiv.org/abs/2610.01351", "title": "Is Success All You Need? Investigating the Impact of Input Perturbations on VLA Behaviour in Tabletop Manipulation Tasks", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.01351 · 2026-10-01
Signals: closed-loop success reported
Vision-Language-Action (VLA) models have achieved high task success rates on robot manipulation task benchmarks.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.01351 · 2026-10-01\nSignals: closed-loop success reported\n\nVision-Language-Action (VLA) models have achieved high task success rates on robot manipulation task benchmarks.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Vision-Language-Action (VLA) models have achieved high task success rates on robot manipulation task benchmarks.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.01351" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.01351", "url": "https://arxiv.org/abs/2610.01351", "title": "Is Success All You Need? Investigating the Impact of Input Perturbations on VLA Behaviour in Tabletop Manipulation Tasks", "publisher": "arXiv 2610.01351", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "Vision-Language-Action (VLA) models have achieved high task success rates on robot manipulation task benchmarks." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.01083", "url": "https://arxiv.org/abs/2610.01083", "title": "WBAG: A Whole-Body and Attached-Geometry Safety Framework for Vision-Language-Action Manipulation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.01083 · 2026-10-01
Signals: closed-loop success reported
Vision-language-action (VLA) policies have demonstrated impressive capabilities in generalizable robotic manipulation, but their deployment in the real world remains challenging due to potential collisions involving different parts of the…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.01083 · 2026-10-01\nSignals: closed-loop success reported\n\nVision-language-action (VLA) policies have demonstrated impressive capabilities in generalizable robotic manipulation, but their deployment in the real world remains challenging due to potential collisions involving different parts of the…\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Vision-language-action (VLA) policies have demonstrated impressive capabilities in generalizable robotic manipulation, but their deployment in the real world remains challenging due to potential collisions involving different parts of the…", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.01083" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.01083", "url": "https://arxiv.org/abs/2610.01083", "title": "WBAG: A Whole-Body and Attached-Geometry Safety Framework for Vision-Language-Action Manipulation", "publisher": "arXiv 2610.01083", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "Vision-language-action (VLA) policies have demonstrated impressive capabilities in generalizable robotic manipulation, but their deployment in the real world remains challenging due to potential collisions involving different parts of the…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00981", "url": "https://arxiv.org/abs/2610.00981", "title": "NarrativeFlow: Flow-Based Vision-Language-Action Model Using Robot Velocity Fields", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.00981 · 2026-10-01
Signals: closed-loop success reported · real-robot experiments
We focus on language-conditioned flow-based manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic, motion-centric representations for leveraging data collected from multiple robot platforms.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.00981 · 2026-10-01\nSignals: closed-loop success reported · real-robot experiments\n\nWe focus on language-conditioned flow-based manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic, motion-centric representations for leveraging data collected from multiple robot platforms.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "We focus on language-conditioned flow-based manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic, motion-centric representations for leveraging data collected from multiple robot platforms.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.00981" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.00981", "url": "https://arxiv.org/abs/2610.00981", "title": "NarrativeFlow: Flow-Based Vision-Language-Action Model Using Robot Velocity Fields", "publisher": "arXiv 2610.00981", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot" ], "excerpt": "We focus on language-conditioned flow-based manipulation, where robot flows (robot velocity fields) serve as embodiment-agnostic, motion-centric representations for leveraging data collected from multiple robot platforms." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00982", "url": "https://arxiv.org/abs/2610.00982", "title": "Divide-and-Remember: Recursive Action-Relevant Memory for Long-Horizon VLA Policies", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.00982 · 2026-10-01
Signals: closed-loop success reported · open code or weights
Vision-language-action (VLA) models struggle on history-dependent manipulation tasks, where the current observation alone does not determine the action, and the policy needs a memory of the history.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.00982 · 2026-10-01\nSignals: closed-loop success reported · open code or weights\n\nVision-language-action (VLA) models struggle on history-dependent manipulation tasks, where the current observation alone does not determine the action, and the policy needs a memory of the history.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Vision-language-action (VLA) models struggle on history-dependent manipulation tasks, where the current observation alone does not determine the action, and the policy needs a memory of the history.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.00982" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.00982", "url": "https://arxiv.org/abs/2610.00982", "title": "Divide-and-Remember: Recursive Action-Relevant Memory for Long-Horizon VLA Policies", "publisher": "arXiv 2610.00982", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "open_release" ], "excerpt": "Vision-language-action (VLA) models struggle on history-dependent manipulation tasks, where the current observation alone does not determine the action, and the policy needs a memory of the history." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.01260", "url": "https://arxiv.org/abs/2610.01260", "title": "PROMO: Preference-conditioned Multi-Objective Reinforcement Learning for Quadrupedal Robots", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.01260 · 2026-10-01
Key numbers: 30.4% · 38.7% · 59.0%
Signals: quantitative results · open code or weights
Quadrupedal locomotion requires balancing conflicting objectives such as command tracking, stability, and energy efficiency, yet conventional reinforcement learning (RL) hardcodes these priorities into a fixed scalar reward at training…\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.01260 · 2026-10-01\nKey numbers: 30.4% · 38.7% · 59.0%\nSignals: quantitative results · open code or weights\n\nQuadrupedal locomotion requires balancing conflicting objectives such as command tracking, stability, and energy efficiency, yet conventional reinforcement learning (RL) hardcodes these priorities into a fixed scalar reward at training…\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Quadrupedal locomotion requires balancing conflicting objectives such as command tracking, stability, and energy efficiency, yet conventional reinforcement learning (RL) hardcodes these priorities into a fixed scalar reward at training…", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.01260" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.01260", "url": "https://arxiv.org/abs/2610.01260", "title": "PROMO: Preference-conditioned Multi-Objective Reinforcement Learning for Quadrupedal Robots", "publisher": "arXiv 2610.01260", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [ "30.4%", "38.7%", "59.0%" ], "signals": [ "numbers", "open_release" ], "excerpt": "Quadrupedal locomotion requires balancing conflicting objectives such as command tracking, stability, and energy efficiency, yet conventional reinforcement learning (RL) hardcodes these priorities into a fixed scalar reward at training…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00899", "url": "https://arxiv.org/abs/2610.00899", "title": "TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.00899 · 2026-10-01
Signals: closed-loop success reported
Autoregressive Vision-Language-Action models often represent continuous robot actions as discrete token sequences, enabling action prediction with standard next-token objectives.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.00899 · 2026-10-01\nSignals: closed-loop success reported\n\nAutoregressive Vision-Language-Action models often represent continuous robot actions as discrete token sequences, enabling action prediction with standard next-token objectives.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Autoregressive Vision-Language-Action models often represent continuous robot actions as discrete token sequences, enabling action prediction with standard next-token objectives.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.00899" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.00899", "url": "https://arxiv.org/abs/2610.00899", "title": "TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models", "publisher": "arXiv 2610.00899", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "Autoregressive Vision-Language-Action models often represent continuous robot actions as discrete token sequences, enabling action prediction with standard next-token objectives." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.02054", "url": "https://arxiv.org/abs/2610.02054", "title": "UniWAM: Unified World-Action Model", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [R]
Source: arXiv 2610.02054 · 2026-10-01
Vision-language-action models benefit from the understanding and reasoning capabilities of pretrained vision-language models, but action-only supervision provides limited grounding in world dynamics.\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [R]\nSource: arXiv 2610.02054 · 2026-10-01\n\nVision-language-action models benefit from the understanding and reasoning capabilities of pretrained vision-language models, but action-only supervision provides limited grounding in world dynamics.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "Vision-language-action models benefit from the understanding and reasoning capabilities of pretrained vision-language models, but action-only supervision provides limited grounding in world dynamics.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.02054" } ], "tags": [ "Data engine (collection / curation / scaling)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.02054", "url": "https://arxiv.org/abs/2610.02054", "title": "UniWAM: Unified World-Action Model", "publisher": "arXiv 2610.02054", "source_id": "arxiv", "evidence": "R", "lane": "data", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [], "excerpt": "Vision-language-action models benefit from the understanding and reasoning capabilities of pretrained vision-language models, but action-only supervision provides limited grounding in world dynamics." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.02196", "url": "https://arxiv.org/abs/2610.02196", "title": "InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation", "content_html": "
Lane: Systems & orchestration (long-horizon / multi-policy)
Evidence: [R]
Source: arXiv 2610.02196 · 2026-10-01
We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining.\n", "content_text": "Lane: Systems & orchestration (long-horizon / multi-policy)\nEvidence: [R]\nSource: arXiv 2610.02196 · 2026-10-01\n\nWe study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining.\n\nFull radar for 2026-10-05: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-05.en.html", "summary": "We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining.", "date_published": "2026-10-04T23:40:00Z", "authors": [ { "name": "arXiv 2610.02196" } ], "tags": [ "Systems & orchestration (long-horizon / multi-policy)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-05", "id": "arxiv:2610.02196", "url": "https://arxiv.org/abs/2610.02196", "title": "InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation", "publisher": "arXiv 2610.02196", "source_id": "arxiv", "evidence": "R", "lane": "systems", "published": "2026-10-01", "radar_published": "2026-10-04T23:40:00Z", "numbers": [], "signals": [], "excerpt": "We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.01612", "url": "https://arxiv.org/abs/2610.01612", "title": "ReCo: Response-Consistent Locomotion with Policy-Aware MPC for Legged Manipulation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.01612 · 2026-10-01
Key numbers: 28.7% · 27.4%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Continuous legged manipulation requires accurate end-effector tracking while the base keeps walking. Combining reinforcement learning (RL) with model predictive control (MPC) suits this task: the learned policy provides robust locomotion,…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.01612 · 2026-10-01\nKey numbers: 28.7% · 27.4%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nContinuous legged manipulation requires accurate end-effector tracking while the base keeps walking. Combining reinforcement learning (RL) with model predictive control (MPC) suits this task: the learned policy provides robust locomotion,…\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Continuous legged manipulation requires accurate end-effector tracking while the base keeps walking. Combining reinforcement learning (RL) with model predictive control (MPC) suits this task: the learned policy provides robust locomotion,…", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2610.01612" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2610.01612", "url": "https://arxiv.org/abs/2610.01612", "title": "ReCo: Response-Consistent Locomotion with Policy-Aware MPC for Legged Manipulation", "publisher": "arXiv 2610.01612", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-10-01", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "28.7%", "27.4%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Continuous legged manipulation requires accurate end-effector tracking while the base keeps walking. Combining reinforcement learning (RL) with model predictive control (MPC) suits this task: the learned policy provides robust locomotion,…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.02161", "url": "https://arxiv.org/abs/2610.02161", "title": "DuoMind: Enabling Distributed Multi-Robot Coordination with Semantic Communication", "content_html": "
Lane: Systems & orchestration (long-horizon / multi-policy)
Evidence: [R]
Source: arXiv 2610.02161 · 2026-10-01
Signals: closed-loop success reported
Vision-language models (VLMs) and vision-language-action models (VLAs) have recently driven rapid progress in general-purpose robots, yet most progress has focused on single-robot settings.\n", "content_text": "Lane: Systems & orchestration (long-horizon / multi-policy)\nEvidence: [R]\nSource: arXiv 2610.02161 · 2026-10-01\nSignals: closed-loop success reported\n\nVision-language models (VLMs) and vision-language-action models (VLAs) have recently driven rapid progress in general-purpose robots, yet most progress has focused on single-robot settings.\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Vision-language models (VLMs) and vision-language-action models (VLAs) have recently driven rapid progress in general-purpose robots, yet most progress has focused on single-robot settings.", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2610.02161" } ], "tags": [ "Systems & orchestration (long-horizon / multi-policy)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2610.02161", "url": "https://arxiv.org/abs/2610.02161", "title": "DuoMind: Enabling Distributed Multi-Robot Coordination with Semantic Communication", "publisher": "arXiv 2610.02161", "source_id": "arxiv", "evidence": "R", "lane": "systems", "published": "2026-10-01", "radar_published": "2026-10-03T23:40:00Z", "numbers": [], "signals": [ "closed_loop" ], "excerpt": "Vision-language models (VLMs) and vision-language-action models (VLAs) have recently driven rapid progress in general-purpose robots, yet most progress has focused on single-robot settings." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.01856", "url": "https://arxiv.org/abs/2610.01856", "title": "ChunkVLA-AM: Parallel Action Chunking for Vision-Language-Action Robot Control in Additive Manufacturing", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2610.01856 · 2026-10-01
Key numbers: 92.9%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Vision-language-action (VLA) models unify visual perception, language understanding, and action generation, offering new opportunities for automation in additive manufacturing (AM).\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2610.01856 · 2026-10-01\nKey numbers: 92.9%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nVision-language-action (VLA) models unify visual perception, language understanding, and action generation, offering new opportunities for automation in additive manufacturing (AM).\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Vision-language-action (VLA) models unify visual perception, language understanding, and action generation, offering new opportunities for automation in additive manufacturing (AM).", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2610.01856" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2610.01856", "url": "https://arxiv.org/abs/2610.01856", "title": "ChunkVLA-AM: Parallel Action Chunking for Vision-Language-Action Robot Control in Additive Manufacturing", "publisher": "arXiv 2610.01856", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-10-01", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "92.9%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Vision-language-action (VLA) models unify visual perception, language understanding, and action generation, offering new opportunities for automation in additive manufacturing (AM)." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39403", "url": "https://arxiv.org/abs/2609.39403", "title": "IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining", "content_html": "
Lane: Data engine (collection / curation / scaling)
Evidence: [R]
Source: arXiv 2609.39403 · 2026-09-30
Key numbers: 10,000 hours · 55.0% · 11.7% · 5,000 hours
Signals: quantitative results · closed-loop success reported
Egocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and…\n", "content_text": "Lane: Data engine (collection / curation / scaling)\nEvidence: [R]\nSource: arXiv 2609.39403 · 2026-09-30\nKey numbers: 10,000 hours · 55.0% · 11.7% · 5,000 hours\nSignals: quantitative results · closed-loop success reported\n\nEgocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and…\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Egocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and…", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2609.39403" } ], "tags": [ "Data engine (collection / curation / scaling)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2609.39403", "url": "https://arxiv.org/abs/2609.39403", "title": "IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining", "publisher": "arXiv 2609.39403", "source_id": "arxiv", "evidence": "R", "lane": "data", "published": "2026-09-30", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "10,000 hours", "55.0%", "11.7%", "5,000 hours" ], "signals": [ "numbers", "closed_loop" ], "excerpt": "Egocentric human video offers a scalable data source for dexterous manipulation, yet using it to train humanoid robots presents two challenges: (1) an embodiment gap, as human hands differ structurally from robot end-effectors and…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.38905", "url": "https://arxiv.org/abs/2609.38905", "title": "EmbodiRSI: Recursive Self-Improvement for Data-Efficient Robot Adaptation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2609.38905 · 2026-09-30
Key numbers: 50.4% · 83.5% · 83.1% · 75.0%
Signals: quantitative results · real-robot experiments
Adapting robot manipulation policies to new tasks and environments remains highly data-intensive, while the data needed for further improvement depends on the policy's current capabilities and failure modes.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2609.38905 · 2026-09-30\nKey numbers: 50.4% · 83.5% · 83.1% · 75.0%\nSignals: quantitative results · real-robot experiments\n\nAdapting robot manipulation policies to new tasks and environments remains highly data-intensive, while the data needed for further improvement depends on the policy's current capabilities and failure modes.\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Adapting robot manipulation policies to new tasks and environments remains highly data-intensive, while the data needed for further improvement depends on the policy's current capabilities and failure modes.", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2609.38905" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2609.38905", "url": "https://arxiv.org/abs/2609.38905", "title": "EmbodiRSI: Recursive Self-Improvement for Data-Efficient Robot Adaptation", "publisher": "arXiv 2609.38905", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-09-30", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "50.4%", "83.5%", "83.1%", "75.0%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Adapting robot manipulation policies to new tasks and environments remains highly data-intensive, while the data needed for further improvement depends on the policy's current capabilities and failure modes." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39611", "url": "https://arxiv.org/abs/2609.39611", "title": "Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2609.39611 · 2026-09-30
Key numbers: 40% · 57% · 300 ms
Signals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers
Agile multi-UAV flight requires accurate and low-latency onboard estimation of the kinematic states of neighboring UAVs for collision avoidance, motion coordination, etc.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2609.39611 · 2026-09-30\nKey numbers: 40% · 57% · 300 ms\nSignals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nAgile multi-UAV flight requires accurate and low-latency onboard estimation of the kinematic states of neighboring UAVs for collision avoidance, motion coordination, etc.\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Agile multi-UAV flight requires accurate and low-latency onboard estimation of the kinematic states of neighboring UAVs for collision avoidance, motion coordination, etc.", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2609.39611" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2609.39611", "url": "https://arxiv.org/abs/2609.39611", "title": "Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation", "publisher": "arXiv 2609.39611", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-09-30", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "40%", "57%", "300 ms" ], "signals": [ "numbers", "closed_loop", "real_robot", "latency" ], "excerpt": "Agile multi-UAV flight requires accurate and low-latency onboard estimation of the kinematic states of neighboring UAVs for collision avoidance, motion coordination, etc." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39600", "url": "https://arxiv.org/abs/2609.39600", "title": "GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2609.39600 · 2026-09-30
Key numbers: 72.42% · 71.35% · 61.75% · 53.32%
Signals: quantitative results · real-robot experiments · latency or control-rate numbers
Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2609.39600 · 2026-09-30\nKey numbers: 72.42% · 71.35% · 61.75% · 53.32%\nSignals: quantitative results · real-robot experiments · latency or control-rate numbers\n\nAutoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens.\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens.", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2609.39600" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2609.39600", "url": "https://arxiv.org/abs/2609.39600", "title": "GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed", "publisher": "arXiv 2609.39600", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-09-30", "radar_published": "2026-10-03T23:40:00Z", "numbers": [ "72.42%", "71.35%", "61.75%", "53.32%" ], "signals": [ "numbers", "real_robot", "latency" ], "excerpt": "Autoregressive (AR) grounding models serialize spatial predictions, introducing sequential latency and imposing a causal order on output tokens." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.38873", "url": "https://arxiv.org/abs/2609.38873", "title": "DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [R]
Source: arXiv 2609.38873 · 2026-09-30
Signals: real-robot experiments
Robots operating in human-centered environments must safely navigate among multiple dynamic obstacles to avoid collisions with people and surrounding infrastructure.\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [R]\nSource: arXiv 2609.38873 · 2026-09-30\nSignals: real-robot experiments\n\nRobots operating in human-centered environments must safely navigate among multiple dynamic obstacles to avoid collisions with people and surrounding infrastructure.\n\nFull radar for 2026-10-04: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-04.en.html", "summary": "Robots operating in human-centered environments must safely navigate among multiple dynamic obstacles to avoid collisions with people and surrounding infrastructure.", "date_published": "2026-10-03T23:40:00Z", "authors": [ { "name": "arXiv 2609.38873" } ], "tags": [ "Safety, permissions & compliance", "[R]" ], "language": "en", "_radar": { "date": "2026-10-04", "id": "arxiv:2609.38873", "url": "https://arxiv.org/abs/2609.38873", "title": "DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles", "publisher": "arXiv 2609.38873", "source_id": "arxiv", "evidence": "R", "lane": "safety", "published": "2026-09-30", "radar_published": "2026-10-03T23:40:00Z", "numbers": [], "signals": [ "real_robot" ], "excerpt": "Robots operating in human-centered environments must safely navigate among multiple dynamic obstacles to avoid collisions with people and surrounding infrastructure." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00575", "url": "https://arxiv.org/abs/2610.00575", "title": "Token-World: World Modeling in Vision-Language Model Token Space for Robot Manipulation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2610.00575 · 2026-09-30
Signals: closed-loop success reported · latency or control-rate numbers
A common approach to world-model simulation for vision-language-action (VLA) systems is to predict future RGB observations and then re-encode them into policy inputs, introducing an indirect interface between simulation and downstream…\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2610.00575 · 2026-09-30\nSignals: closed-loop success reported · latency or control-rate numbers\n\nA common approach to world-model simulation for vision-language-action (VLA) systems is to predict future RGB observations and then re-encode them into policy inputs, introducing an indirect interface between simulation and downstream…\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "A common approach to world-model simulation for vision-language-action (VLA) systems is to predict future RGB observations and then re-encode them into policy inputs, introducing an indirect interface between simulation and downstream…", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "arXiv 2610.00575" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "arxiv:2610.00575", "url": "https://arxiv.org/abs/2610.00575", "title": "Token-World: World Modeling in Vision-Language Model Token Space for Robot Manipulation", "publisher": "arXiv 2610.00575", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-09-30", "radar_published": "2026-10-02T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "latency" ], "excerpt": "A common approach to world-model simulation for vision-language-action (VLA) systems is to predict future RGB observations and then re-encode them into policy inputs, introducing an indirect interface between simulation and downstream…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00360", "url": "https://arxiv.org/abs/2610.00360", "title": "DexPolicy: Scheduled Exploration for Trajectory-Guided Dexterous Manipulation", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2610.00360 · 2026-09-30
Key numbers: 49.4% · 68.1% · 14.1% · 45.4%
Signals: quantitative results · closed-loop success reported · open code or weights
Reinforcement learning (RL) for dexterous manipulation must discover finger-object contacts and then control the object precisely; the action noise that serves the first goal can interfere with the second.\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2610.00360 · 2026-09-30\nKey numbers: 49.4% · 68.1% · 14.1% · 45.4%\nSignals: quantitative results · closed-loop success reported · open code or weights\n\nReinforcement learning (RL) for dexterous manipulation must discover finger-object contacts and then control the object precisely; the action noise that serves the first goal can interfere with the second.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "Reinforcement learning (RL) for dexterous manipulation must discover finger-object contacts and then control the object precisely; the action noise that serves the first goal can interfere with the second.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "arXiv 2610.00360" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "arxiv:2610.00360", "url": "https://arxiv.org/abs/2610.00360", "title": "DexPolicy: Scheduled Exploration for Trajectory-Guided Dexterous Manipulation", "publisher": "arXiv 2610.00360", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-09-30", "radar_published": "2026-10-02T23:40:00Z", "numbers": [ "49.4%", "68.1%", "14.1%", "45.4%" ], "signals": [ "numbers", "closed_loop", "open_release" ], "excerpt": "Reinforcement learning (RL) for dexterous manipulation must discover finger-object contacts and then control the object precisely; the action noise that serves the first goal can interfere with the second." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00864", "url": "https://arxiv.org/abs/2610.00864", "title": "Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.00864 · 2026-10-01
Key numbers: 67.5% · 74.4% · 30.3% · 54.9%
Signals: quantitative results · open code or weights · latency or control-rate numbers
In this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.00864 · 2026-10-01\nKey numbers: 67.5% · 74.4% · 30.3% · 54.9%\nSignals: quantitative results · open code or weights · latency or control-rate numbers\n\nIn this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "In this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "arXiv 2610.00864" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "arxiv:2610.00864", "url": "https://arxiv.org/abs/2610.00864", "title": "Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models", "publisher": "arXiv 2610.00864", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-10-01", "radar_published": "2026-10-02T23:40:00Z", "numbers": [ "67.5%", "74.4%", "30.3%", "54.9%" ], "signals": [ "numbers", "open_release", "latency" ], "excerpt": "In this paper, we study how to achieve one-step action generation in Robotic Foundation Models (RFMs), aiming to overcome the high inference latency of multi-step flow matching." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2610.00355", "url": "https://arxiv.org/abs/2610.00355", "title": "IndoorBEV: A Lightweight Real-Time LiDAR BEV Perception System for Indoor Mobile Robots", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2610.00355 · 2026-09-30
Key numbers: 0.6M parameters · 169.6 ms · 200 ms
Signals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers
Efficient indoor LiDAR perception is challenging because mobile robots must understand cluttered three-dimensional environments under strict latency and memory constraints.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2610.00355 · 2026-09-30\nKey numbers: 0.6M parameters · 169.6 ms · 200 ms\nSignals: quantitative results · real-robot experiments · open code or weights · latency or control-rate numbers\n\nEfficient indoor LiDAR perception is challenging because mobile robots must understand cluttered three-dimensional environments under strict latency and memory constraints.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "Efficient indoor LiDAR perception is challenging because mobile robots must understand cluttered three-dimensional environments under strict latency and memory constraints.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "arXiv 2610.00355" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "arxiv:2610.00355", "url": "https://arxiv.org/abs/2610.00355", "title": "IndoorBEV: A Lightweight Real-Time LiDAR BEV Perception System for Indoor Mobile Robots", "publisher": "arXiv 2610.00355", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-09-30", "radar_published": "2026-10-02T23:40:00Z", "numbers": [ "0.6M parameters", "169.6 ms", "200 ms" ], "signals": [ "numbers", "real_robot", "open_release", "latency" ], "excerpt": "Efficient indoor LiDAR perception is challenging because mobile robots must understand cluttered three-dimensional environments under strict latency and memory constraints." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:aws-physical-ai:b9f6fd764f44", "url": "https://aws.amazon.com/blogs/physical-ai/isaac-lab-on-aws-from-simulation-to-registered-policy/", "title": "Isaac Lab on AWS: From Simulation to Registered Policy", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [O]
Source: AWS Physical AI Blog · 2026-10-02
Signals: real-robot experiments
Introduction: The missing layer in Physical AI on AWS Physical AI systems, those that perceive, reason, and act in the physical world, are moving fast from research into production deployments.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [O]\nSource: AWS Physical AI Blog · 2026-10-02\nSignals: real-robot experiments\n\nIntroduction: The missing layer in Physical AI on AWS Physical AI systems, those that perceive, reason, and act in the physical world, are moving fast from research into production deployments.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "Introduction: The missing layer in Physical AI on AWS Physical AI systems, those that perceive, reason, and act in the physical world, are moving fast from research into production deployments.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "AWS Physical AI Blog" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[O]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "aws-physical-ai:b9f6fd764f44", "url": "https://aws.amazon.com/blogs/physical-ai/isaac-lab-on-aws-from-simulation-to-registered-policy/", "title": "Isaac Lab on AWS: From Simulation to Registered Policy", "publisher": "AWS Physical AI Blog", "source_id": "aws-physical-ai", "evidence": "O", "lane": "simeval", "published": "2026-10-02", "radar_published": "2026-10-02T23:40:00Z", "numbers": [], "signals": [ "real_robot" ], "excerpt": "Introduction: The missing layer in Physical AI on AWS Physical AI systems, those that perceive, reason, and act in the physical world, are moving fast from research into production deployments." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:ieee-spectrum-robotics:00bb782fa200", "url": "https://spectrum.ieee.org/video-friday-bioinspired-robotics", "title": "Video Friday: Albatross Falls, Spins, Self-Rights, and Sails Away", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: IEEE Spectrum Robotics · 2026-10-02
We’ve developed ALBATROSS, a hybrid aerial–marine robot that can be released from the air, autorotate down to the water without a parachute, passively self-right after landing, and then reuse the same rigid wings as sails for autonomous…\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: IEEE Spectrum Robotics · 2026-10-02\n\nWe’ve developed ALBATROSS, a hybrid aerial–marine robot that can be released from the air, autorotate down to the water without a parachute, passively self-right after landing, and then reuse the same rigid wings as sails for autonomous…\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "We’ve developed ALBATROSS, a hybrid aerial–marine robot that can be released from the air, autorotate down to the water without a parachute, passively self-right after landing, and then reuse the same rigid wings as sails for autonomous…", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "IEEE Spectrum Robotics" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "ieee-spectrum-robotics:00bb782fa200", "url": "https://spectrum.ieee.org/video-friday-bioinspired-robotics", "title": "Video Friday: Albatross Falls, Spins, Self-Rights, and Sails Away", "publisher": "IEEE Spectrum Robotics", "source_id": "ieee-spectrum-robotics", "evidence": "M", "lane": "hardware", "published": "2026-10-02", "radar_published": "2026-10-02T23:40:00Z", "numbers": [], "signals": [], "excerpt": "We’ve developed ALBATROSS, a hybrid aerial–marine robot that can be released from the air, autorotate down to the water without a parachute, passively self-right after landing, and then reuse the same rigid wings as sails for autonomous…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:013398c4f77b", "url": "https://www.therobotreport.com/runway-introduces-praxis-1-world-action-model-robotics/", "title": "Runway introduces Praxis-1 world action model for robotics", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [M]
Source: The Robot Report · 2026-10-02
Signals: open code or weights
When Runway does release Praxis-1 publicly, it plans to ship it with open weights rather than as a closed model.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [M]\nSource: The Robot Report · 2026-10-02\nSignals: open code or weights\n\nWhen Runway does release Praxis-1 publicly, it plans to ship it with open weights rather than as a closed model.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "When Runway does release Praxis-1 publicly, it plans to ship it with open weights rather than as a closed model.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Foundation models (VLA / WAM)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "the-robot-report:013398c4f77b", "url": "https://www.therobotreport.com/runway-introduces-praxis-1-world-action-model-robotics/", "title": "Runway introduces Praxis-1 world action model for robotics", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "foundation", "published": "2026-10-02", "radar_published": "2026-10-02T23:40:00Z", "numbers": [], "signals": [ "open_release" ], "excerpt": "When Runway does release Praxis-1 publicly, it plans to ship it with open weights rather than as a closed model." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:71461f9591ea", "url": "https://www.therobotreport.com/inside-omrons-next-generation-ld-mobile-robots/", "title": "Inside Omron’s next-generation LD mobile robots", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [M]
Source: The Robot Report · 2026-10-02
In mobile robotics, software, safety, and fleet management are becoming just as important as the robot itself, according to an OMROM project manager.\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [M]\nSource: The Robot Report · 2026-10-02\n\nIn mobile robotics, software, safety, and fleet management are becoming just as important as the robot itself, according to an OMROM project manager.\n\nFull radar for 2026-10-03: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-03.en.html", "summary": "In mobile robotics, software, safety, and fleet management are becoming just as important as the robot itself, according to an OMROM project manager.", "date_published": "2026-10-02T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Safety, permissions & compliance", "[M]" ], "language": "en", "_radar": { "date": "2026-10-03", "id": "the-robot-report:71461f9591ea", "url": "https://www.therobotreport.com/inside-omrons-next-generation-ld-mobile-robots/", "title": "Inside Omron’s next-generation LD mobile robots", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "safety", "published": "2026-10-02", "radar_published": "2026-10-02T23:40:00Z", "numbers": [], "signals": [], "excerpt": "In mobile robotics, software, safety, and fleet management are becoming just as important as the robot itself, according to an OMROM project manager." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39198", "url": "https://arxiv.org/abs/2609.39198", "title": "DSDyn-VLA: A Dual-Stream Dynamic Manipulation Framework with Motion Perception, Future Awareness, and Realtime Correction", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2609.39198 · 2026-09-30
Signals: closed-loop success reported · real-robot experiments · open code or weights · latency or control-rate numbers
While Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation).\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2609.39198 · 2026-09-30\nSignals: closed-loop success reported · real-robot experiments · open code or weights · latency or control-rate numbers\n\nWhile Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation).\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "While Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation).", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "arXiv 2609.39198" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "arxiv:2609.39198", "url": "https://arxiv.org/abs/2609.39198", "title": "DSDyn-VLA: A Dual-Stream Dynamic Manipulation Framework with Motion Perception, Future Awareness, and Realtime Correction", "publisher": "arXiv 2609.39198", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-09-30", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "real_robot", "open_release", "latency" ], "excerpt": "While Vision-Language-Action (VLA) models excel in static tasks, they struggle in dynamic environments where objects are in motion (e.g., conveyor belt manipulation)." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39763", "url": "https://arxiv.org/abs/2609.39763", "title": "DiffWAM: A Fast and Efficient Navigation World Action Model", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2609.39763 · 2026-09-30
Key numbers: 74.40%
Signals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers
Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2609.39763 · 2026-09-30\nKey numbers: 74.40%\nSignals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nPretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "arXiv 2609.39763" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "arxiv:2609.39763", "url": "https://arxiv.org/abs/2609.39763", "title": "DiffWAM: A Fast and Efficient Navigation World Action Model", "publisher": "arXiv 2609.39763", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-09-30", "radar_published": "2026-10-01T23:40:00Z", "numbers": [ "74.40%" ], "signals": [ "numbers", "closed_loop", "real_robot", "latency" ], "excerpt": "Pretrained video foundation models encode rich semantic and spatiotemporal priors for embodied navigation, yet converting these priors into UAV motion typically requires expensive future-video synthesis and geometric reconstruction." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.37771", "url": "https://arxiv.org/abs/2609.37771", "title": "Faster and Better? Benchmark Bugs and Design Limitations Distort the Evaluation of Vision-Language-Action Acceleration", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2609.37771 · 2026-09-29
Signals: closed-loop success reported · open code or weights · latency or control-rate numbers
Simulated manipulation benchmarks are the standard tool for evaluating vision-language-action (VLA) policies and the acceleration methods that reduce their inference latency for on-robot deployment.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2609.37771 · 2026-09-29\nSignals: closed-loop success reported · open code or weights · latency or control-rate numbers\n\nSimulated manipulation benchmarks are the standard tool for evaluating vision-language-action (VLA) policies and the acceleration methods that reduce their inference latency for on-robot deployment.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "Simulated manipulation benchmarks are the standard tool for evaluating vision-language-action (VLA) policies and the acceleration methods that reduce their inference latency for on-robot deployment.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "arXiv 2609.37771" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "arxiv:2609.37771", "url": "https://arxiv.org/abs/2609.37771", "title": "Faster and Better? Benchmark Bugs and Design Limitations Distort the Evaluation of Vision-Language-Action Acceleration", "publisher": "arXiv 2609.37771", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-09-29", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [ "closed_loop", "open_release", "latency" ], "excerpt": "Simulated manipulation benchmarks are the standard tool for evaluating vision-language-action (VLA) policies and the acceleration methods that reduce their inference latency for on-robot deployment." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.39822", "url": "https://arxiv.org/abs/2609.39822", "title": "Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation", "content_html": "
Lane: Foundation models (VLA / WAM)
Evidence: [R]
Source: arXiv 2609.39822 · 2026-09-30
Key numbers: 61.557 ms · 21.956 ms
Signals: quantitative results · closed-loop success reported · latency or control-rate numbers
Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution.\n", "content_text": "Lane: Foundation models (VLA / WAM)\nEvidence: [R]\nSource: arXiv 2609.39822 · 2026-09-30\nKey numbers: 61.557 ms · 21.956 ms\nSignals: quantitative results · closed-loop success reported · latency or control-rate numbers\n\nVision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "arXiv 2609.39822" } ], "tags": [ "Foundation models (VLA / WAM)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "arxiv:2609.39822", "url": "https://arxiv.org/abs/2609.39822", "title": "Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation", "publisher": "arXiv 2609.39822", "source_id": "arxiv", "evidence": "R", "lane": "foundation", "published": "2026-09-30", "radar_published": "2026-10-01T23:40:00Z", "numbers": [ "61.557 ms", "21.956 ms" ], "signals": [ "numbers", "closed_loop", "latency" ], "excerpt": "Vision-language-action (VLA) models face a timing gap between low-rate inference and high-rate robot execution." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:ieee-spectrum-robotics:73878f37956b", "url": "https://spectrum.ieee.org/robust-robot-hand", "title": "Atlas Robot’s New Hand May Outperform Humanlike Designs", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: IEEE Spectrum Robotics · 2026-10-01
Today, Boston Dynamics is announcing a redesigned hand for its Atlas robot . The previous generation could do some crazy superhuman things with its three fingers, but it was never intended for mass production in the tens or hundreds of…\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: IEEE Spectrum Robotics · 2026-10-01\n\nToday, Boston Dynamics is announcing a redesigned hand for its Atlas robot . The previous generation could do some crazy superhuman things with its three fingers, but it was never intended for mass production in the tens or hundreds of…\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "Today, Boston Dynamics is announcing a redesigned hand for its Atlas robot . The previous generation could do some crazy superhuman things with its three fingers, but it was never intended for mass production in the tens or hundreds of…", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "IEEE Spectrum Robotics" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "ieee-spectrum-robotics:73878f37956b", "url": "https://spectrum.ieee.org/robust-robot-hand", "title": "Atlas Robot’s New Hand May Outperform Humanlike Designs", "publisher": "IEEE Spectrum Robotics", "source_id": "ieee-spectrum-robotics", "evidence": "M", "lane": "hardware", "published": "2026-10-01", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [], "excerpt": "Today, Boston Dynamics is announcing a redesigned hand for its Atlas robot . The previous generation could do some crazy superhuman things with its three fingers, but it was never intended for mass production in the tens or hundreds of…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:147d35dd5178", "url": "https://www.therobotreport.com/boston-dynamics-drops-pinkie-on-new-humanoid-hand/", "title": "Boston Dynamics drops pinkie on new humanoid hand", "content_html": "
Lane: Embodiment & supply chain (hardware / cost)
Evidence: [M]
Source: The Robot Report · 2026-10-01
Boston Dynamics engineers taped their pinkies and ring fingers together for a day to learn what they weren’t able to do.\n", "content_text": "Lane: Embodiment & supply chain (hardware / cost)\nEvidence: [M]\nSource: The Robot Report · 2026-10-01\n\nBoston Dynamics engineers taped their pinkies and ring fingers together for a day to learn what they weren’t able to do.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "Boston Dynamics engineers taped their pinkies and ring fingers together for a day to learn what they weren’t able to do.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Embodiment & supply chain (hardware / cost)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "the-robot-report:147d35dd5178", "url": "https://www.therobotreport.com/boston-dynamics-drops-pinkie-on-new-humanoid-hand/", "title": "Boston Dynamics drops pinkie on new humanoid hand", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "hardware", "published": "2026-10-01", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [], "excerpt": "Boston Dynamics engineers taped their pinkies and ring fingers together for a day to learn what they weren’t able to do." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:99805b279c40", "url": "https://www.therobotreport.com/your-robots-safety-functions-already-work-what-if-the-input-lies/", "title": "Your Robot’s Safety Functions Already Work. What If the Input Lies?", "content_html": "
Lane: Safety, permissions & compliance
Evidence: [M]
Source: The Robot Report · 2026-10-01
A robot can follow its safety rules and still act on manipulated information. See what this means for robot safety and how teams can strengthen their testing and evidence.\n", "content_text": "Lane: Safety, permissions & compliance\nEvidence: [M]\nSource: The Robot Report · 2026-10-01\n\nA robot can follow its safety rules and still act on manipulated information. See what this means for robot safety and how teams can strengthen their testing and evidence.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "A robot can follow its safety rules and still act on manipulated information. See what this means for robot safety and how teams can strengthen their testing and evidence.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Safety, permissions & compliance", "[M]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "the-robot-report:99805b279c40", "url": "https://www.therobotreport.com/your-robots-safety-functions-already-work-what-if-the-input-lies/", "title": "Your Robot’s Safety Functions Already Work. What If the Input Lies?", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "safety", "published": "2026-10-01", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [], "excerpt": "A robot can follow its safety rules and still act on manipulated information. See what this means for robot safety and how teams can strengthen their testing and evidence." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:the-robot-report:7d547966b72a", "url": "https://www.therobotreport.com/anybotics-launches-shift-streamline-robot-fleet-operations-scale-autonomous-inspections/", "title": "ANYbotics launches Shift to streamline robot fleet operations, scale inspections", "content_html": "
Lane: Systems & orchestration (long-horizon / multi-policy)
Evidence: [M]
Source: The Robot Report · 2026-10-01
ANYbotics designed Shift to help maintenance teams run industrial inspections, detect failing equipment, and turn data into work orders.\n", "content_text": "Lane: Systems & orchestration (long-horizon / multi-policy)\nEvidence: [M]\nSource: The Robot Report · 2026-10-01\n\nANYbotics designed Shift to help maintenance teams run industrial inspections, detect failing equipment, and turn data into work orders.\n\nFull radar for 2026-10-02: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-02.en.html", "summary": "ANYbotics designed Shift to help maintenance teams run industrial inspections, detect failing equipment, and turn data into work orders.", "date_published": "2026-10-01T23:40:00Z", "authors": [ { "name": "The Robot Report" } ], "tags": [ "Systems & orchestration (long-horizon / multi-policy)", "[M]" ], "language": "en", "_radar": { "date": "2026-10-02", "id": "the-robot-report:7d547966b72a", "url": "https://www.therobotreport.com/anybotics-launches-shift-streamline-robot-fleet-operations-scale-autonomous-inspections/", "title": "ANYbotics launches Shift to streamline robot fleet operations, scale inspections", "publisher": "The Robot Report", "source_id": "the-robot-report", "evidence": "M", "lane": "systems", "published": "2026-10-01", "radar_published": "2026-10-01T23:40:00Z", "numbers": [], "signals": [], "excerpt": "ANYbotics designed Shift to help maintenance teams run industrial inspections, detect failing equipment, and turn data into work orders." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.38059", "url": "https://arxiv.org/abs/2609.38059", "title": "WorldLine: Action-Driven Visual Simulation for Robotic Manipulation", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2609.38059 · 2026-09-29
Key numbers: 10,000 hours · 2,000 hours · 74%
Signals: quantitative results · closed-loop success reported · real-robot experiments
Real-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2609.38059 · 2026-09-29\nKey numbers: 10,000 hours · 2,000 hours · 74%\nSignals: quantitative results · closed-loop success reported · real-robot experiments\n\nReal-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors.\n\nFull radar for 2026-10-01: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-01.en.html", "summary": "Real-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors.", "date_published": "2026-09-30T23:40:00Z", "authors": [ { "name": "arXiv 2609.38059" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-01", "id": "arxiv:2609.38059", "url": "https://arxiv.org/abs/2609.38059", "title": "WorldLine: Action-Driven Visual Simulation for Robotic Manipulation", "publisher": "arXiv 2609.38059", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-09-29", "radar_published": "2026-09-30T23:40:00Z", "numbers": [ "10,000 hours", "2,000 hours", "74%" ], "signals": [ "numbers", "closed_loop", "real_robot" ], "excerpt": "Real-world robot learning is constrained by the cost of collecting experience and evaluating candidate behaviors." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.37560", "url": "https://arxiv.org/abs/2609.37560", "title": "RoboFin3D: A Sim-to-Real Platform for Robotic Surface Finishing", "content_html": "
Lane: Simulation & evaluation (sim2real / closed-loop)
Evidence: [R]
Source: arXiv 2609.37560 · 2026-09-29
Key numbers: 77.15% · 84.47% · 97.41%
Signals: quantitative results · real-robot experiments
Grinding and sanding are fundamental processes in industrial robotic surface finishing. However, physical trials are expensive and consume workpieces, making reproducible experiments difficult.\n", "content_text": "Lane: Simulation & evaluation (sim2real / closed-loop)\nEvidence: [R]\nSource: arXiv 2609.37560 · 2026-09-29\nKey numbers: 77.15% · 84.47% · 97.41%\nSignals: quantitative results · real-robot experiments\n\nGrinding and sanding are fundamental processes in industrial robotic surface finishing. However, physical trials are expensive and consume workpieces, making reproducible experiments difficult.\n\nFull radar for 2026-10-01: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-01.en.html", "summary": "Grinding and sanding are fundamental processes in industrial robotic surface finishing. However, physical trials are expensive and consume workpieces, making reproducible experiments difficult.", "date_published": "2026-09-30T23:40:00Z", "authors": [ { "name": "arXiv 2609.37560" } ], "tags": [ "Simulation & evaluation (sim2real / closed-loop)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-01", "id": "arxiv:2609.37560", "url": "https://arxiv.org/abs/2609.37560", "title": "RoboFin3D: A Sim-to-Real Platform for Robotic Surface Finishing", "publisher": "arXiv 2609.37560", "source_id": "arxiv", "evidence": "R", "lane": "simeval", "published": "2026-09-29", "radar_published": "2026-09-30T23:40:00Z", "numbers": [ "77.15%", "84.47%", "97.41%" ], "signals": [ "numbers", "real_robot" ], "excerpt": "Grinding and sanding are fundamental processes in industrial robotic surface finishing. However, physical trials are expensive and consume workpieces, making reproducible experiments difficult." } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.36588", "url": "https://arxiv.org/abs/2609.36588", "title": "Cooperative Multi-Agent Vision-Language-Action Models via Reinforced Fine Tuning", "content_html": "
Lane: Training & self-improvement (RL / post-training)
Evidence: [R]
Source: arXiv 2609.36588 · 2026-09-29
Key numbers: 11 tasks
Signals: closed-loop success reported · real-robot experiments
We study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained…\n", "content_text": "Lane: Training & self-improvement (RL / post-training)\nEvidence: [R]\nSource: arXiv 2609.36588 · 2026-09-29\nKey numbers: 11 tasks\nSignals: closed-loop success reported · real-robot experiments\n\nWe study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained…\n\nFull radar for 2026-10-01: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-01.en.html", "summary": "We study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained…", "date_published": "2026-09-30T23:40:00Z", "authors": [ { "name": "arXiv 2609.36588" } ], "tags": [ "Training & self-improvement (RL / post-training)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-01", "id": "arxiv:2609.36588", "url": "https://arxiv.org/abs/2609.36588", "title": "Cooperative Multi-Agent Vision-Language-Action Models via Reinforced Fine Tuning", "publisher": "arXiv 2609.36588", "source_id": "arxiv", "evidence": "R", "lane": "training", "published": "2026-09-29", "radar_published": "2026-09-30T23:40:00Z", "numbers": [ "11 tasks" ], "signals": [ "closed_loop", "real_robot" ], "excerpt": "We study reinforcement learning (RL) methods for cooperative multi-agent Vision-Language-Action (VLA) models. This problem is challenging because VLAs are pretrained on large-scale single-agent data and therefore lack the fine-grained…" } }, { "id": "tag:noteflowai.github.io,2026:physical-ai-radar:arxiv:2609.37772", "url": "https://arxiv.org/abs/2609.37772", "title": "Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control", "content_html": "
Lane: Edge & real-time (on-device inference / control rate)
Evidence: [R]
Source: arXiv 2609.37772 · 2026-09-29
Key numbers: 1.89x
Signals: closed-loop success reported · real-robot experiments · latency or control-rate numbers
Diffusion and flow-matching Vision-Language-Action (VLA) policies generate action chunks through iterative denoising, incurring substantial inference latency that severely limits real-time robotic control.\n", "content_text": "Lane: Edge & real-time (on-device inference / control rate)\nEvidence: [R]\nSource: arXiv 2609.37772 · 2026-09-29\nKey numbers: 1.89x\nSignals: closed-loop success reported · real-robot experiments · latency or control-rate numbers\n\nDiffusion and flow-matching Vision-Language-Action (VLA) policies generate action chunks through iterative denoising, incurring substantial inference latency that severely limits real-time robotic control.\n\nFull radar for 2026-10-01: https://noteflowai.github.io/physical-ai-radar/radar/daily/2026-10-01.en.html", "summary": "Diffusion and flow-matching Vision-Language-Action (VLA) policies generate action chunks through iterative denoising, incurring substantial inference latency that severely limits real-time robotic control.", "date_published": "2026-09-30T23:40:00Z", "authors": [ { "name": "arXiv 2609.37772" } ], "tags": [ "Edge & real-time (on-device inference / control rate)", "[R]" ], "language": "en", "_radar": { "date": "2026-10-01", "id": "arxiv:2609.37772", "url": "https://arxiv.org/abs/2609.37772", "title": "Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control", "publisher": "arXiv 2609.37772", "source_id": "arxiv", "evidence": "R", "lane": "edge", "published": "2026-09-29", "radar_published": "2026-09-30T23:40:00Z", "numbers": [ "1.89x" ], "signals": [ "closed_loop", "real_robot", "latency" ], "excerpt": "Diffusion and flow-matching Vision-Language-Action (VLA) policies generate action chunks through iterative denoising, incurring substantial inference latency that severely limits real-time robotic control." } } ] }