Physical AI Radar · 2026-10-13

Language: ZH · EN · JA

A daily radar on the Physical AI frontier: eight lanes, trilingual digests, traceable evidence.

• Generated: 2026-10-12 23:40 UTC
• Window: papers 2026-10-09 → 2026-10-13 · posts 2026-09-13 → 2026-10-13 · no repeats within 30 days (UTC)
• Evidence tags: O = first-party official · R = paper/preprint · M = media/secondary

Today’s picks

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

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [R] ｜ Source: arXiv 2610.13643 ｜ 2026-10-09
• Key numbers: 14 tasks (“AutoBio+ comprises 14 tasks across 4 robot”)
• Signals: closed-loop success reported · real-robot experiments
• Why it matters AI draft: If you are automating lab bench work, this preprint offers 14 tasks across four robots that test dexterous tool use under realistic visual conditions and dynamic experimental phenomena; real-robot and closed-loop signals are flagged, but results are self-reported.

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.

2. Forward Dynamics Models

• Lane: Foundation models (VLA / WAM) ｜ Evidence: [R] ｜ Source: arXiv 2610.13117 ｜ 2026-10-09
• Key numbers: 99.4% (“achieving a state-of-the-art 99.4% average success rate”) · 13% (“gains of up to 13% over VLA and”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments
• Why it matters AI draft: When choosing between a VLA and a world action model as your policy backbone, this preprint argues for forward dynamics models, self-reporting a 99.4% average success rate and gains of up to 13% over VLA baselines; the page does not name the benchmark.

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.

3. Evo-RL: Iterative Robot Reinforcement Learning from Selective Experience

• Lane: Training & self-improvement (RL / post-training) ｜ Evidence: [R] ｜ Source: arXiv 2610.13345 ｜ 2026-10-09
• Key numbers: 1,000 hours (“which spans over 1,000 hours across four robot”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights
• Why it matters AI draft: If your real-world RL loop mixes human corrections with autonomous rollouts, this preprint proposes iterating on selectively chosen experience, cites over 1,000 hours across four robots and flags an open release, so you can test it on your own platform before trusting the claims.

Existing real-world reinforcement learning methods improve robot policies by combining human corrections with autonomous rollouts.

4. 3D Gaussian Splatting in practice: Industry use cases

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [O] ｜ Source: AWS Physical AI Blog ｜ 2026-10-06
• Why it matters AI draft: If photorealistic scene capture is the bottleneck in your simulation pipeline, AWS argues Gaussian Splatting has cut the cost and complexity of photorealistic 3D enough to make near real-time capture, robotics and simulation viable use cases; it is a vendor position with no figures and no sim-to-real evidence on the page.

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…

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

• Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence: [M] ｜ Source: The Robot Report ｜ 2026-10-12
• Signals: latency or control-rate numbers
• Why it matters AI draft: Operators setting KPIs for automated systems may note that Daifuku’s Cecile Dick-Calmes, as quoted, no longer judges success solely by automation density or equipment utilization; it is a secondary opinion piece and the page gives no real-time figures.

Daifuku’s Cecile Dick-Calmes said success is no longer measured solely by automation density or equipment utilization.

6. 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

• Lane: Embodiment & supply chain (hardware / cost) ｜ Evidence: [M] ｜ Source: 雷峰网 Leiphone ｜ 2026-10-12
• Why it matters AI draft: If you are weighing whether map-free driving techniques carry over to robots, read this as one founder’s view from outside both robotics labs and the recent Physical AI startup wave; it is a profile, with no data or deployment evidence on the page.

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。

7. 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

• Lane: Data engine (collection / curation / scaling) ｜ Evidence: [M] ｜ Source: 雷峰网 Leiphone ｜ 2026-10-12
• Why it matters AI draft: Humanoid teams looking for third-party evaluation on real job tasks can track the review round in Shanghai on October 23–25, framed under an MIIT and SASAC real-scenario training campaign; this is a secondary preview with no results yet.

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。

8. 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」

• Lane: Training & self-improvement (RL / post-training) ｜ Evidence: [M] ｜ Source: 雷峰网 Leiphone ｜ 2026-10-12
• Why it matters AI draft: There is nothing yet for robot builders to act on: the excerpt gives only the event date, the Beijing venue and ticketing, with no robotics or training content named, so wait for the agenda before treating it as a signal.

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）。

Frontier baseline (curated)

1. Egocentric human video can outperform real-robot data for embodied pretraining

• Lane: Data engine (collection / curation / scaling) ｜ Evidence: [R] ｜ Source: arXiv 2606.20521 ｜ 2026
• Key numbers: -24% validation loss at equal data · +52.5% in-distribution success · +90% out-of-distribution success
• Why it matters: Teams without a fleet can still build a data advantage; teams with a fleet but no curation and judging pipeline hold data that is only a storage cost.

2. Fine-tuning π0 for robotic manipulation on Amazon SageMaker HyperPod EKS

• Lane: Training & self-improvement (RL / post-training) ｜ Evidence: [O] ｜ Source: AWS Physical AI Blog ｜ 2026-09-10
• Key numbers: MSE -89.7% (DROID) · MSE -88.9% (LIBERO) · 8xH100, 20k steps, 6h15m · 50-step chunk <200 ms at 1 ODE step · checkpoint ~45 GB
• Signals: quantitative results
• Why it matters: An earlier version of the same work reported ~96% because eval episodes overlapped training; enforcing the split moved it to ~89%. The best evaluation-discipline case of 2026.

3. RoboWorld: fast and reliable neural simulators for generalist robot policy evaluation

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [R] ｜ Source: arXiv 2607.01060 ｜ 2026
• Key numbers: closed-loop rollouts in a video world model · VLM judge · rankings strongly correlated with real-world eval
• Why it matters: If rankings inside a world model align with real-robot rankings, the evaluation budget shifts from human hours to GPU hours - real-robot spot checks still cannot be dropped.

4. Bringing a frontier world model to the convenience store: Telexistence’s DreamZero experiment on AWS

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [O] ｜ Source: AWS Physical AI Blog ｜ 2026-09-09
• Key numbers: 14B world action model · best closed-loop 45% · real-data-trained run 0% in sim · MSE -55.6% from 1k to 2k steps · ~583 ms/step on 8xB200 · curation flagged 39.3% vs human 49.7%
• Signals: quantitative results
• Why it matters: Turns ‘is a frontier model right for my robot’ into measured numbers, and honestly records the 0% when real data was evaluated in sim - the domain-gap diagnosis is itself the deliverable.

5. EgoScale: scaling dexterous manipulation with diverse egocentric human data

• Lane: Data engine (collection / curation / scaling) ｜ Evidence: [R] ｜ Source: arXiv 2602.16710 ｜ 2026
• Key numbers: 44k hours of egocentric video · +54% average success vs no-pretraining on a 22-DoF hand · transfers to lower-DoF hands
• Why it matters: Large-scale human motion is shown to be a reusable, embodiment-agnostic motor prior, so dexterous hands do not start from zero.

6. RoboTTT: context scaling for robot policies via test-time training

• Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence: [R] ｜ Source: arXiv 2607.15275 (NVIDIA) ｜ 2026-07-16
• Key numbers: 8K timestep visuomotor context · 3 orders of magnitude beyond prior policies · no inference-latency growth
• Signals: latency or control-rate numbers
• Why it matters: Long context enters the policy without added latency, so one-shot in-context imitation and on-the-fly improvement can live in the same inference pass.

Lane distribution (picks, last 7 days)

Recent cadence

• Source mix (picks, last 7 days): [O] 1 · [R] 36 · [M] 19
• Picks per recent run: 10-07 8 · 10-08 8 · 10-09 8 · 10-10 8 · 10-11 8 · 10-12 8 · 10-13 8

Standards & compliance watch

• ISO 25785-1 — Safety requirements for dynamically stable industrial mobile robots — Under development; committee-stage as of 2026-09.
• Regulation (EU) 2023/1230 (Machinery Regulation) — Fully applicable 2027-01-20.
• EU AI Act official timeline (Council of the EU) — High-risk deadlines 2027-12-02 / 2028-08-02 per Digital Omnibus.

Per-item analysis on this page was drafted by radar-analyst (claude-opus-5.5) and merged automatically after deterministic checks and an agent review, without human review; titles and numbers are untouched.

This repository is technical observation. It is not investment advice, a product commitment, or a safety certification conclusion. Paper claims are author-reported.

Methodology · Weekly roundup · Archive