Physical AI Radar · 2026-10-11
A daily radar on the Physical AI frontier: eight lanes, trilingual digests, traceable evidence.
- Generated:
2026-10-10 23:40 UTC - Window:
papers 2026-10-07 → 2026-10-11 · posts 2026-09-11 → 2026-10-11 · no repeats within 30 days (UTC) - Evidence tags: O = first-party official · R = paper/preprint · M = media/secondary
Today’s picks
1. RESETTLE: Robotic Recovery through Disagreement-Triggered Retrieval and Efficient Corrective Control
- Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence:
[R]｜ Source: arXiv 2610.12185 ｜2026-10-08 - Key numbers:
8.70%(“RESETTLE achieves up to 8.70%”) ·6.28%·6.83%(“and 6.83% absolute success-rate gains”) ·74.04%(“monitoring-and-recovery computation latency is 74.04% –93.57% lower than”) - Signals:
quantitative results·real-robot experiments·open code or weights·latency or control-rate numbers - Why it matters
AI draft: If your manipulation stack lacks timely correction after execution errors, this self-reported preprint claims success-rate gains of up to 8.70% with lower monitoring-and-recovery latency; the page does not name the comparison, so check it before adopting.
Reliable robotic manipulation requires timely intervention to correct emerging deviations and restore progress after execution errors.
2. Being-M0.7: A Latent World-Action Model for Humanoid Robots
- Lane: Embodiment & supply chain (hardware / cost) ｜ Evidence:
[R]｜ Source: arXiv 2610.11283 ｜2026-10-08 - Key numbers:
10,000 hours(“corpus from more than 10,000 hours of raw human-centric”) - Signals:
quantitative results·closed-loop success reported·real-robot experiments - Why it matters
AI draft: For humanoid loco-manipulation teams short of robot demonstrations, this preprint builds on more than 10,000 hours of raw human-centric data; results are author-reported, so transfer to your own platform still needs your own testing.
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.
3. Residual Modeling Closes the Regression and Generative Policy Gap in Robot Learning
- Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence:
[R]｜ Source: arXiv 2610.12231 ｜2026-10-08 - Signals:
closed-loop success reported·real-robot experiments - Why it matters
AI draft: If you are choosing between diffusion or flow-matching policies and plain MSE regression, this preprint is worth reading for its closed-loop and real-robot experiments, but the page shows no figures to justify switching architectures yet.
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…
4. ARC: A Reasoning Recipe for Robot Foundation Models
- Lane: Foundation models (VLA / WAM) ｜ Evidence:
[R]｜ Source: arXiv 2610.12386 ｜2026-10-08 - Signals:
closed-loop success reported·real-robot experiments - Why it matters
AI draft: Teams planning to improve a robot foundation model through larger models, more demonstrations and costly training get a reasoning-centred alternative to weigh; it is a self-reported preprint with closed-loop and real-robot signals but no figures on the page.
The prevailing approach to improving robot foundation models (RFMs) relies on larger models, more robot demonstrations, and costly training at scale.
5. SkillWeave: Weaving Heterogeneous Demonstrations into Long-Horizon Manipulation Skills
- Lane: Data engine (collection / curation / scaling) ｜ Evidence:
[R]｜ Source: arXiv 2610.12046 ｜2026-10-08 - Key numbers:
27%(“SkillWeave achieves 27% average end-to-end success”) ·65%(“to an average of 65%”) ·87%(“average composition efficiency of 87%”) - Signals:
quantitative results·closed-loop success reported·real-robot experiments - Why it matters
AI draft: If your dexterous demonstrations must cover both long task progression and contact-rich interaction, this preprint reports 27% average end-to-end success and 87% composition efficiency; the end-to-end figure is modest and self-reported, so reproduce before reshaping collection.
Dexterous manipulation requires both large-scale task progression and precise contact-rich interaction, making it challenging to collect demonstrations that effectively support both regimes.
6. FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems
- Lane: Safety, permissions & compliance ｜ Evidence:
[R]｜ Source: arXiv 2610.12432 ｜2026-10-08 - Key numbers:
99.95%(“it reaches a 99.95% safety rate while”) ·97%(“safety rate while retaining 97% of the unfiltered”) - Signals:
quantitative results·real-robot experiments - Why it matters
AI draft: If your safe RL puts safety and task performance in one objective that pulls both ways, this preprint reports a 99.95% safety rate while retaining 97% relative to the unfiltered policy; it is self-reported, and a safety rate is not a certification argument.
Safe reinforcement learning commonly places safety and task performance in the same policy objective, where they can introduce competing updates.
7. Rewiring Semantics, Dynamics, and Control: A Simple yet Effective Action-Centric Tri-Stream Transformer
- Lane: Foundation models (VLA / WAM) ｜ Evidence:
[R]｜ Source: arXiv 2610.11416 ｜2026-10-08 - Signals:
real-robot experiments - Why it matters
AI draft: For VLAs built on a pretrained VLM’s semantic understanding, this preprint proposes an action-centric architecture and reports real-robot experiments; with no figures on the page, it is one to track rather than a reason to change your model stack now.
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).
8. Why humanoid robot demos still fail the generalization test
- Lane: Embodiment & supply chain (hardware / cost) ｜ Evidence:
[M]｜ Source: The Robot Report ｜2026-10-09 - Signals:
real-robot experiments - Why it matters
AI draft: When assessing a humanoid vendor’s demo, this secondary piece argues that moving beyond demos to reliable real-world performance depends on better data and human judgment; it gives no figures, so use it to frame procurement questions, not as evidence.
How better data and human judgment can help robots move beyond demos toward reliable real-world performance.
Frontier baseline (curated)
1. PL-CBF: finite-horizon safety at runtime via parallel fallback rollouts
- Lane: Safety, permissions & compliance ｜ Evidence:
[R]｜ Source: arXiv 2605.16588 ｜2026 - Key numbers:
library of fallback policies·least-invasive safe mode·QP minimally modifies the nominal policy - Why it matters: ‘Least-invasive fallback’ can go straight into customer acceptance criteria - far more concrete than ‘we are safe’.
2. RD-VLA: implicit test-time compute scaling via latent iterative reasoning
- Lane: Edge & real-time (on-device inference / control rate) ｜ Evidence:
[R]｜ Source: arXiv 2602.07845 ｜2026 - Key numbers:
constant memory regardless of reasoning depth·up to 80x inference speedup vs token-based reasoning VLAs - Signals:
latency or control-rate numbers - Why it matters: Robots can ‘think longer’ without paying token latency - the key to moving test-time compute inside the control loop.
3. ISO 25785-1 (under development): safety for dynamically stable industrial mobile robots
- Lane: Safety, permissions & compliance ｜ Evidence:
[O]｜ Source: ISO ｜2026 - Key numbers:
committee stage as of 2026-09·covers legged / self-balancing robots - Why it matters: Humanoids break the fixed-base premise of ISO 10218; until the dedicated standard lands, safety rests on architecture and runtime guarantees.
4. GR00T N1.7: open reasoning VLA model for humanoid robots
- Lane: Foundation models (VLA / WAM) ｜ Evidence:
[O]｜ Source: NVIDIA / Hugging Face ｜2026 - Key numbers:
3B parameters·open weights - Why it matters: A 3B open reasoning VLA gives in-house teams a fine-tunable, deployable starting point instead of building a foundation model.
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)
| Lane | # |
|---|---|
| Foundation models (VLA / WAM) | 12 |
| Data engine (collection / curation / scaling) | 5 |
| Training & self-improvement (RL / post-training) | 4 |
| Simulation & evaluation (sim2real / closed-loop) | 14 |
| Edge & real-time (on-device inference / control rate) | 6 |
| Safety, permissions & compliance | 2 |
| Systems & orchestration (long-horizon / multi-policy) | 3 |
| Embodiment & supply chain (hardware / cost) | 10 |
Recent cadence
- Source mix (picks, last 7 days):
[O]0 ·[R]39 ·[M]17 - Picks per recent run: 10-05 8 · 10-06 8 · 10-07 8 · 10-08 8 · 10-09 8 · 10-10 8 · 10-11 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.