Physical AI Radar

AutoBio+: Benchmarking Dexterous Manipulation with Dynamic Experimental Phenomena in Realistic Laboratories

2026-10-12T23:40:00Z

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.

Full radar for 2026-10-13

Forward Dynamics Models

2026-10-12T23:40:00Z

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.

Full radar for 2026-10-13

Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience

2026-10-12T23:40:00Z

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.

Full radar for 2026-10-13

3D Gaussian Splatting in practice: Industry use cases

2026-10-12T23:40:00Z

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…

Full radar for 2026-10-13

Closing the reality gap: Why real-time execution has become the new competitive advantage

2026-10-12T23:40:00Z

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.

Full radar for 2026-10-13

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

2026-10-12T23:40:00Z

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。

Full radar for 2026-10-13

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

2026-10-12T23:40:00Z

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。

Full radar for 2026-10-13

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」

2026-10-12T23:40:00Z

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）。

Full radar for 2026-10-13

LiteNWM: Efficient Latent World Models for Onboard Visual Navigation in the Wild

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic Manipulation

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors

2026-10-11T23:40:00Z

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…

Full radar for 2026-10-12

CSF: Contextual Safety Filtering for Motion Generators

2026-10-11T23:40:00Z

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.

Full radar for 2026-10-12

RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

Being-M0.7: A Latent World-Action Model for Humanoid Robots

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning

2026-10-10T23:40:00Z

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…

Full radar for 2026-10-11

ARC: A Reasoning Recipe for Robot Foundation Models

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer

2026-10-10T23:40:00Z

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).

Full radar for 2026-10-11

Why humanoid robot demos still fail the generalization test

2026-10-10T23:40:00Z

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.

Full radar for 2026-10-11

RoboRender: Robot-Oriented Video Generation for Visual Sim-to-Real Transfer

2026-10-10T07:15:00Z

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.

Full radar for 2026-10-10

USDCraft: Geometrically Grounded Programmatic Modeling of Articulated 3D Assets for Simulation

2026-10-10T07:15:00Z

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.

Full radar for 2026-10-10

Humanoid World Action Model With Joint State--Action Generation

2026-10-10T07:15:00Z

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…

Full radar for 2026-10-10

ManiUnit: A Manipulation Skill Dataset and Benchmark for Long-Horizon Tasks

2026-10-10T07:15:00Z

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.

Full radar for 2026-10-10

Video Friday: Robot Decommissioning Takes a Fun Turn

2026-10-10T07:15:00Z

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.

Full radar for 2026-10-10

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

2026-10-10T07:15:00Z

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。

Full radar for 2026-10-10

这届 IROS 2026 上，具身学术与产业边界正在消融

2026-10-10T07:15:00Z

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”。

Full radar for 2026-10-10

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

2026-10-10T07:15:00Z

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！

Full radar for 2026-10-10

WARP-VLA: Wrist-Camera Adaptation for View-Robust Policy Execution in Vision-Language-Action Models

2026-10-09T07:28:00Z

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.

Full radar for 2026-10-09

REACT: Rolling Denoising and Dual Decoupling for Reactive Robot Control with VLA Models

2026-10-09T07:28:00Z

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…

Full radar for 2026-10-09

Leveraging Human-In-The-Loop Demonstrations in Reinforcement Learning for Digital Twin-Driven Robot Flexibility

2026-10-09T07:28:00Z

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.

Full radar for 2026-10-09

Tell Robot What Not to Do: A Negation Understanding Perspective

2026-10-09T07:28:00Z

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.

Full radar for 2026-10-09

This Disembodied Hand Is All the Robot You Need

2026-10-09T07:28:00Z

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?

Full radar for 2026-10-09

Jabil discusses the pace of humanoid robot development and production

2026-10-09T07:28:00Z

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.

Full radar for 2026-10-09

AWS launches open-source Physical AI Toolchain for robotics

2026-10-09T07:28:00Z

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.

Full radar for 2026-10-09

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

2026-10-09T07:28:00Z

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）。

Full radar for 2026-10-09

Agentic RSR: Real-to-Sim-to-Real through Scene Reconstruction and Execution-Grounded Robot Policies

2026-10-08T07:30:00Z

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.

Full radar for 2026-10-08

RealtimeWAM: How Fast Can I Run My World Action Model?

2026-10-08T07:30:00Z

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.

Full radar for 2026-10-08

RFPO: Rectified Flow Policy Optimization for Embodied Control

2026-10-08T07:30:00Z

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.

Full radar for 2026-10-08

PEARS: Physical-Prior-Guided Efficient Adaptation via Failure Reasoning and Diffusion Steering for Tactile Manipulation

2026-10-08T07:30:00Z

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.

Full radar for 2026-10-08

研究了1933篇 IROS 论文，我们看到了机器人学的六项新变化

2026-10-08T07:30:00Z

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。

Full radar for 2026-10-08

对话如祺出行COO韩锋：世界模型越强，真实数据才是真壁垒

2026-10-08T07:30:00Z

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。

Full radar for 2026-10-08

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

2026-10-08T07:30:00Z

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。

Full radar for 2026-10-08

Harmonic Drive to share how integrated actuators can reduce engineering at RoboBusiness

2026-10-08T07:30:00Z

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.

Full radar for 2026-10-08

Demo: Closed-Loop Sionna-Isaac Sim Co-Simulation Framework for Wireless-Aware Robot Navigation over ROS 2

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

StairVLA: Stage-Aware Hierarchical Action Generation for Vision-Language-Action Models

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

ESP: Energy-Score Policy for One-Step Multimodal Action Generation

2026-10-07T07:19:00Z

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…

Full radar for 2026-10-07

Adapting Vision-Language-Action Models to Unknown Visual Disruptions During Execution

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

LightVLN: Efficient Aerial Vision-and-Language Navigation with Compact Memory and History-Guided Local Aggregation

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

ActTune: Action-Aware Precision and GPU Operating-Point Adaptation for Energy-Efficient Vision-Language-Action Inference

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

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

2026-10-07T07:19:00Z

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。

Full radar for 2026-10-07

Boston Dynamics appoints former Amazon exec Rohit Prasad as new CEO

2026-10-07T07:19:00Z

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.

Full radar for 2026-10-07

FastOPD: On-Policy Distillation for Lightweight VLA Deployment

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

UniIntervene++: An Adaptive Intervention Agent for Efficient Real-World Reinforcement Learning

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

MobiAgent: Dual-Loop Recursive Policy Self-Improvement for Long-Horizon Mobile Manipulation

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

World Action Learning via Interaction-Centric Spectral Latent Guidance

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

SimpleTouch: Can Vision-Language-Action Models Master Contact-Rich Manipulation Without Tactile Policy Pretraining?

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

HexVIO: Towards All-Day Stereo-Inertial Tracking Through Commodity DSPs

2026-10-05T23:40:00Z

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.

Full radar for 2026-10-06

FCC robot restrictions could accelerate shift to local AI

2026-10-05T23:40:00Z

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…

Full radar for 2026-10-06

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

2026-10-05T23:40:00Z

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。

Full radar for 2026-10-06

Is Success All You Need? Investigating the Impact of Input Perturbations on VLA Behaviour in Tabletop Manipulation Tasks

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

WBAG: A Whole-Body and Attached-Geometry Safety Framework for Vision-Language-Action Manipulation

2026-10-04T23:40:00Z

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…

Full radar for 2026-10-05

NarrativeFlow: Flow-Based Vision-Language-Action Model Using Robot Velocity Fields

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

Divide-and-Remember: Recursive Action-Relevant Memory for Long-Horizon VLA Policies

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

PROMO: Preference-conditioned Multi-Objective Reinforcement Learning for Quadrupedal Robots

2026-10-04T23:40:00Z

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…

Full radar for 2026-10-05

TOAST: Stochastic Robot Action Tokenization for Autoregressive Vision-Language-Action Models

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

UniWAM: Unified World-Action Model

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation

2026-10-04T23:40:00Z

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.

Full radar for 2026-10-05

ReCo: Response-Consistent Locomotion with Policy-Aware MPC for Legged Manipulation

2026-10-03T23:40:00Z

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,…

Full radar for 2026-10-04

DuoMind: Enabling Distributed Multi-Robot Coordination with Semantic Communication

2026-10-03T23:40:00Z

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.

Full radar for 2026-10-04

ChunkVLA-AM: Parallel Action Chunking for Vision-Language-Action Robot Control in Additive Manufacturing

2026-10-03T23:40:00Z

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).

Full radar for 2026-10-04

IronMind: Scaling Humanoid Dexterous Manipulation via Camera-Space Ego-Centric Pretraining

2026-10-03T23:40:00Z

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…

Full radar for 2026-10-04

EmbodiRSI: Recursive Self-Improvement for Data-Efficient Robot Adaptation

2026-10-03T23:40:00Z

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.

Full radar for 2026-10-04

Towards Agile Vision-Based Multi-UAV Flight: Revisiting State Estimation

2026-10-03T23:40:00Z

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.

Full radar for 2026-10-04

GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed

2026-10-03T23:40:00Z

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.

Full radar for 2026-10-04

DODGER: Safety-Guided Reinforcement Learning for Robot Navigation Among Dynamic Obstacles

2026-10-03T23:40:00Z

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.

Full radar for 2026-10-04

Token-World: World Modeling in Vision-Language Model Token Space for Robot Manipulation

2026-10-02T23:40:00Z

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…

Full radar for 2026-10-03

DexPolicy: Scheduled Exploration for Trajectory-Guided Dexterous Manipulation

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

Kinematic MeanFlow: One-Step Action Generation Policy for Robotic Foundation Models

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

IndoorBEV: A Lightweight Real-Time LiDAR BEV Perception System for Indoor Mobile Robots

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

Isaac Lab on AWS: From Simulation to Registered Policy

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

Video Friday: Albatross Falls, Spins, Self-Rights, and Sails Away

2026-10-02T23:40:00Z

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…

Full radar for 2026-10-03

Runway introduces Praxis-1 world action model for robotics

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

Inside Omron’s next-generation LD mobile robots

2026-10-02T23:40:00Z

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.

Full radar for 2026-10-03

DSDyn-VLA: A Dual-Stream Dynamic Manipulation Framework with Motion Perception, Future Awareness, and Realtime Correction

2026-10-01T23:40:00Z

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).

Full radar for 2026-10-02

DiffWAM: A Fast and Efficient Navigation World Action Model

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

Faster and Better? Benchmark Bugs and Design Limitations Distort the Evaluation of Vision-Language-Action Acceleration

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

Toward Real-Time VLAs: Stage-Aware Two-Step Flow Denoising and System-Level Evaluation

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

Atlas Robot’s New Hand May Outperform Humanlike Designs

2026-10-01T23:40:00Z

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…

Full radar for 2026-10-02

Boston Dynamics drops pinkie on new humanoid hand

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

Your Robot’s Safety Functions Already Work. What If the Input Lies?

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

ANYbotics launches Shift to streamline robot fleet operations, scale inspections

2026-10-01T23:40:00Z

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.

Full radar for 2026-10-02

WorldLine: Action-Driven Visual Simulation for Robotic Manipulation

2026-09-30T23:40:00Z

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.

Full radar for 2026-10-01

RoboFin3D: A Sim-to-Real Platform for Robotic Surface Finishing

2026-09-30T23:40:00Z

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.

Full radar for 2026-10-01

Cooperative Multi-Agent Vision-Language-Action Models via Reinforced Fine Tuning

2026-09-30T23:40:00Z

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…

Full radar for 2026-10-01

Urgent Actions Go First: Urgency-Aware Denoising for Real-Time VLA Control

2026-09-30T23:40:00Z

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.

Full radar for 2026-10-01