Physical AI Radar · 2026-10-12
A daily radar on the Physical AI frontier: eight lanes, trilingual digests, traceable evidence.
- Generated:
2026-10-11 23:40 UTC - Window:
papers 2026-10-08 → 2026-10-12 · posts 2026-09-12 → 2026-10-12 · no repeats within 30 days (UTC) - Evidence tags: O = first-party official · R = paper/preprint · M = media/secondary
Today’s picks
1. LiteNWM: Efficient Latent World Models for Onboard Visual Navigation in the Wild
- Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence:
[R]｜ Source: arXiv 2610.12368 ｜2026-10-08 - Key numbers:
17.56%(“macro-averaged trajectory error by 17.56% relative to NoMaD+NWM-XL”) ·16.2%(“macro-averaged trajectory error by 16.2%”) ·43.3%(“improves navigation success from 43.3% to 83.3% relative”) ·83.3%(“success from 43.3% to 83.3% relative to NoMaD”) - Signals:
quantitative results·closed-loop success reported·real-robot experiments - Why it matters
AI draft: If your visual navigation policy generates trajectories without evaluating their consequences, this preprint reports navigation success rising from 43.3% to 83.3% relative to NoMaD, with real-robot and closed-loop signals; the figures are self-reported, so retest on your own routes.
Direct visual navigation policies generate trajectories efficiently but do not explicitly evaluate their future consequences.
2. RoboAware: Learning to Coordinate Embodied Skills from Counterfactual Outcomes
- Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence:
[R]｜ Source: arXiv 2610.11480 ｜2026-10-08 - Key numbers:
100 tasks(“Comprehensive single-episode evaluations on 100 tasks show that RoboAware”) ·77.0%(“that RoboAware reaches a 77.0% overall success rate”) ·90.0%(“with SOTA averages of 90.0% on RoboSuite”) ·73.8%(“73.8% on diverse LIBERO-Pro”) - Signals:
quantitative results·closed-loop success reported - Why it matters
AI draft: If a coding agent composes your modular skills with frozen end-to-end policies, this preprint reports a 77.0% overall success rate across 100 tasks; the page shows benchmark results only, with no real-robot evidence, so treat it as a routing idea to test.
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.
3. Generative Neural Retargeting for Human-to-Robot Dexterous Manipulation
- Lane: Data engine (collection / curation / scaling) ｜ Evidence:
[R]｜ Source: arXiv 2610.12440 ｜2026-10-08 - Signals:
closed-loop success reported - Why it matters
AI draft: Teams that want human demonstrations for dexterous manipulation but are blocked by the embodiment gap should track this preprint; the page shows a closed-loop signal but no figures, which is not yet enough to change a collection plan.
Human demonstrations are a scalable data source for learning dexterous manipulation, but the embodiment gap prevents human motion from being executed directly on robots.
4. LeWAM: A JEPA World Action Model with Diffusion-Steering-Based MPC
- Lane: Foundation models (VLA / WAM) ｜ Evidence:
[R]｜ Source: arXiv 2610.12407 ｜2026-10-08 - Signals:
closed-loop success reported - Why it matters
AI draft: If your world action model sits on a reconstruction-based representation, this preprint argues that such representations carry noisy, redundant information that complicates downstream prediction; with a closed-loop signal but no figures on the page, it is a direction, not grounds to switch.
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.
5. SpatialHarness: Test-Time Spatial Scaffolding for Fine Robotic Manipulation
- Lane: Training & self-improvement (RL / post-training) ｜ Evidence:
[R]｜ Source: arXiv 2610.12457 ｜2026-10-08 - Key numbers:
26.7%(“including from 26.7% to 66.7% on”) ·66.7%(“including from 26.7% to 66.7% on plug insertion”) ·0%(“plug insertion and from 0% to 100% on”) ·100%(“and from 0% to 100% on Tower of”) - Signals:
quantitative results·closed-loop success reported·real-robot experiments - Why it matters
AI draft: If you drive a robot directly with a frontier multimodal foundation model and hit limits on fine manipulation, this preprint reports plug insertion going from 26.7% to 66.7%, with real-robot signals; these are self-reported per-task figures, so generality is unproven.
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.
6. Real-Time Motion Planning with Dynamic Hazards: Classical vs. Learning-Based Methods
- 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 - Why it matters
AI draft: When choosing between classical and learning-based motion planners for dynamic hazard fields, this controlled-comparison preprint is useful input; the page carries closed-loop and latency signals but no figures, so read its test conditions before acting on the verdict.
We study real-time motion planning in dynamic hazard fields through a controlled comparison between classical planning and learning-based methods.
7. WAND: Learning Robust Navigation under Complex Wind Disturbances and Dense Obstacles for Quadrotors
- 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 - Why it matters
AI draft: For quadrotor teams flying through clutter where strong wind perturbs dynamics and limits control authority, this preprint is worth tracking; it shows closed-loop and latency signals but no figures on the page, so fit to your airframe still needs testing.
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…
8. CSF: Contextual Safety Filtering for Motion Generators
- Lane: Safety, permissions & compliance ｜ Evidence:
[R]｜ Source: arXiv 2610.12467 ｜2026-10-08 - Key numbers:
90%(“rate by up to 90%”) ·100%(“while preserving 88-100% of benign motions”) - Signals:
quantitative results·real-robot experiments - Why it matters
AI draft: If text-conditioned motion generators drive your whole-body motion, they cannot tell whether an action targets an object or a person; this preprint reports a rate cut by up to 90% with real-robot signals, but the page excerpt does not name the metric, and it is no certification argument.
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.
Frontier baseline (curated)
1. 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.
2. π*0.6 / Recap: a VLA that learns from experience via advantage-conditioned policies
- Lane: Training & self-improvement (RL / post-training) ｜ Evidence:
[R]｜ Source: arXiv 2511.14759 ｜2026 - Key numbers:
RL from autonomous rollouts + human corrections·value function fine-tuned on online data - Why it matters: The methodological basis for ‘deployment is training’: to use it, operations must first make intervention logs and success judgements first-class data.
3. LeRobotDataset v3.0: a standardised, streamable format for robot learning data
- Lane: Data engine (collection / curation / scaling) ｜ Evidence:
[O]｜ Source: Hugging Face ｜2026 - Key numbers:
Parquet + MP4 + relational metadata·multi-episode packing·native streaming for multi-TB datasets - Why it matters: The format question is converging on LeRobot v3 for data and OpenUSD for assets; data you can convert in one step is worth far more than the same hours locked in a private format.
4. Isaac Lab-Arena: reusable APIs to author, scale and analyse robot policy evaluations
- Lane: Simulation & evaluation (sim2real / closed-loop) ｜ Evidence:
[O]｜ Source: NVIDIA ｜2026 - Key numbers:
policy evaluation at scale·solver-agnostic - Why it matters: Evaluation moves from one-off scripts to a reusable pipeline - the engineering piece that turns ‘can demo’ into ‘can regress’.
5. 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.
6. 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.
Lane distribution (picks, last 7 days)
| Lane | # |
|---|---|
| Foundation models (VLA / WAM) | 11 |
| 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) | 7 |
| Safety, permissions & compliance | 3 |
| Systems & orchestration (long-horizon / multi-policy) | 2 |
| Embodiment & supply chain (hardware / cost) | 10 |
Recent cadence
- Source mix (picks, last 7 days):
[O]0 ·[R]39 ·[M]17 - Picks per recent run: 10-06 8 · 10-07 8 · 10-08 8 · 10-09 8 · 10-10 8 · 10-11 8 · 10-12 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.