Physical AI Radar · 2026-10-06

Language: ZH · EN · JA

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

• Generated: 2026-10-05 23:40 UTC
• Window: papers 2026-10-02 → 2026-10-06 · posts 2026-09-06 → 2026-10-06 · no repeats within 30 days (UTC)
• Evidence tags: O = first-party official · R = paper/preprint · M = media/secondary

Today’s picks

1. FastOPD: On-Policy Distillation for Lightweight VLA Deployment

• Lane: Foundation models (VLA / WAM) ｜ Evidence: [R] ｜ Source: arXiv 2610.02832 ｜ 2026-10-02
• Key numbers: 84% (“FastOPD retains 84% of the performance”) · 78.1% (“reducing inference latency by 78.1% while outperforming existing”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments · latency or control-rate numbers
• Why it matters: Preprint, not yet peer-reviewed; the text gives real-robot results, closed-loop results and figures. Watch whether it cuts the data a new task or embodiment needs.

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.

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

• Lane: Training & self-improvement (RL / post-training) ｜ Evidence: [R] ｜ Source: arXiv 2610.03620 ｜ 2026-10-02
• Key numbers: 89.67% (“average success rate of 89.67%”) · 0.77% (“reducing human intervention to 0.77%”) · 94.6% (“reduction of at least 94.6% from the best”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments · open code or weights
• Why it matters: Preprint, not yet peer-reviewed; the text gives real-robot results, closed-loop results and figures. Watch whether the gain holds outside the training tasks.

Online reinforcement learning (RL) enables robot policies to improve through physical interaction, but the assistance they require changes as their competence evolves.

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

• Lane: Systems & orchestration (long-horizon / multi-policy) ｜ Evidence: [R] ｜ Source: arXiv 2610.03476 ｜ 2026-10-02
• Key numbers: 7.50% (“success improves from 7.50% to 27.50% on”) · 27.50% (“improves from 7.50% to 27.50% on RoboCasa and”) · 32.5% (“on RoboCasa and from 32.5% to 57.5% on”) · 57.5% (“and from 32.5% to 57.5% on Astribot S1”)
• Signals: quantitative results · real-robot experiments
• Why it matters: Preprint, not yet peer-reviewed; the text gives no closed-loop results. Watch whether it holds up over long tasks without human resets.

Long-horizon mobile manipulation presents significant challenges due to compounding execution errors and capacity interference between locomotion and arm control.

4. World Action Learning via Interaction-Centric Spectral Latent Guidance

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [R] ｜ Source: arXiv 2610.03607 ｜ 2026-10-02
• Key numbers: 99.20% (“average success rates of 99.20% on LIBERO”) · 93.80% (“93.80% on RoboTwin 2.0”) · 57.7% (“and 57.7% on RoboCasa-GR1”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments
• Why it matters: Preprint, not yet peer-reviewed; the text gives real-robot results, closed-loop results and figures. Watch how well the simulated result predicts real-robot performance.

Learning general-purpose robot policies requires large-scale real-world interaction data, yet collecting robot demonstrations remains expensive and difficult to scale.

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

• Lane: Foundation models (VLA / WAM) ｜ Evidence: [R] ｜ Source: arXiv 2610.02784 ｜ 2026-10-02
• Key numbers: 50 demonstrations (“With 50 demonstrations per task”) · 77.5% (“average success rate reaches 77.5%”) · 45.2% (“compared with 45.2% for FTP-$π_{0.5}$ and”) · 66.7% (“45.2% for FTP-$π_{0.5}$ and 66.7% for FTP-1”)
• Signals: quantitative results · closed-loop success reported · real-robot experiments
• Why it matters: Preprint, not yet peer-reviewed; the text gives real-robot results, closed-loop results and figures. Watch whether it cuts the data a new task or embodiment needs.

Tactile sensing provides essential contact information for robotic manipulation, yet incorporating it into pretrained vision-language-action (VLA) models remains challenging.

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

• Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence: [R] ｜ Source: arXiv 2610.03283 ｜ 2026-10-02
• Key numbers: 67% (“demonstrates a 67% reduction in power”) · 86% (“power consumption or an 86% increase in throughput”) · 30 fps (“to sustain long-term real-time 30 fps tracking for 0.83”) · 0.83 W (“30 fps tracking for 0.83 W”)
• Signals: quantitative results · latency or control-rate numbers
• Why it matters: Preprint, not yet peer-reviewed; the text gives no real-robot results or closed-loop results. Watch whether the speed-up holds on the target hardware at control rate.

The ability of a device to localize itself within its surroundings is a fundamental prerequisite for spatial computing.

7. FCC robot restrictions could accelerate shift to local AI

• Lane: Safety, permissions & compliance ｜ Evidence: [M] ｜ Source: The Robot Report ｜ 2026-10-05
• Why it matters: Secondary report, so check the primary source before citing; the text gives no figures. Watch what it changes for certification and deployment approval.

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…

8. 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

• Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence: [M] ｜ Source: MONOist (ITmedia) ｜ 2026-10-05
• Why it matters: Secondary report, so check the primary source before citing; the text gives no figures. Watch how well the simulated result predicts real-robot performance.

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。

Frontier baseline (curated)

1. XPENG commissions its humanoid production lines; IRON walks off the line autonomously

• Lane: Embodiment & supply chain (hardware / cost) ｜ Evidence: [O] ｜ Source: PR Newswire ｜ 2026-09-08
• Key numbers: >80% of core processes automated (reported) · target 1,000+ units/month by year-end (reported)
• Why it matters: Line commissioning is a verifiable fact; the monthly target is reported second-hand. Separating the two is the easiest citation mistake to get caught on this cycle.

2. HumanEgo: zero-shot robot learning from minutes of human egocentric video

• Lane: Data engine (collection / curation / scaling) ｜ Evidence: [R] ｜ Source: arXiv 2605.24934 ｜ 2026
• Key numbers: 30 min/task -> 92.5% average success on 4 real tasks · 15 min -> 75% · +41% over matched-time teleoperation
• Why it matters: It pushes the cost per new task down to minutes of data - the most direct yardstick for whether your data strategy is behind.

3. Jetson-PI: onboard real-time control via foresight-aligned asynchronous inference

• Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence: [R] ｜ Source: arXiv 2607.12659 ｜ 2026
• Key numbers: 8.66x control frequency vs naive PyTorch on Orin · +14.8% average success vs VLASH on LIBERO
• Signals: latency or control-rate numbers
• Why it matters: Nearly an order of magnitude in control frequency from systems engineering on the same hardware - deployment optimisation often has more headroom than swapping models.

4. Gemini Robotics 2 brings whole body intelligence to robots

• Lane: Foundation models (VLA / WAM) ｜ Evidence: [O] ｜ Source: Google DeepMind ｜ 2026
• Key numbers: full humanoid control, feet to fingertips · bi-arm robots
• Why it matters: Whole-body control enters the generalist VLA feature list, so upper-body manipulation and lower-body balance stop living in two separate systems.

5. 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.

6. NVIDIA Jetson Thor: up to 2070 FP4 TFLOPS and 128 GB at 40-130 W

• Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence: [O] ｜ Source: NVIDIA ｜ 2026
• Key numbers: 7.5x AI compute vs Orin · 3.5x energy efficiency vs AGX Orin · Jetson T2000: 400 FP4 TFLOPS / 16 GB
• Signals: quantitative results · latency or control-rate numbers
• Why it matters: A generational jump in on-device compute creates room for world models and test-time compute, yet the real-time budget still has to be measured item by item.

Lane distribution (picks, last 7 days)

Recent cadence

• Source mix (picks, last 7 days): [O] 1 · [R] 39 · [M] 16
• Picks per recent run: 09-30 8 · 10-01 8 · 10-02 8 · 10-03 8 · 10-04 8 · 10-05 8 · 10-06 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.

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