Physical AI weekly: Isaac Lab 3.0 ships, robot evals expose huge gaps
AIFlow_ML · x · 2026-10-24
AIFlow's Physical AI weekly deep-dive: every layer of the robot stack moved in one week. Key points:
- Data bottleneck: startups attack it with gloves and datasets; Forbes reports Maxinsights raised $30M seed to fix robot training data, with $47.4B flowing into physical AI in H1 2026 — nearly 4x H2 2025.
- NVIDIA productizes the stack: Isaac Lab 3.0 GA, plus Isaac OS and a reference humanoid with Agile Robots.
- Evals arrive: the XDOF chart shows LLM agents hitting 52-86% on retrieval tasks but collapsing on long-horizon manipulation — best overall score just 19%, with folding, knot-tying and stacking at 0%.
Related event: Physical AI Weekly: Isaac Lab 3.0, Haptic Data Gloves and Humanoid Robots(2 posts)→
More from Embodied
- MorphMimic: one shared policy dances across Unitree G1, AgiBot X2 Ultra, Booster T2 — ChongZzZhang · 2026-10-24
- Grok lays out a path to billions of humanoid robots by the 2030s — Dr_Singularity · 2026-10-24
- Contact data is the new moat for dexterity: cloud robot brains vs on-board reflexes — AIFlow_ML · 2026-10-24
- Swedish startup IntuiCell's digital nervous system lets a robot dog learn to stand without pre-training — CurieuxExplorer · 2026-10-24
- Motoniq robot agent hits 92% on perturbed LIBERO tasks, next best 53% — hbouammar · 2026-10-24
- Tesla engineer: Cybercab camera tuning done with Grok, coding fixed-function hardware directly — yunta_tsai · 2026-10-24