HumanCLAW by Meta Reveals SOTA VLMs Are Surprisingly Bad at Embodied Action
liuziwei7 · x · 2026-08-01
HumanCLAW, a collaborative project by Meta, NTU, and UW, tests the "Action Intelligence" of Vision-Language Models (VLMs)—their ability to decide moment-by-moment what a physical body should do next based on an egocentric view and high-level instructions.
The research reveals that today's best models are surprisingly bad at it. The framework decouples decision-making from motor control: a VLM outputs atomic skill decisions, a verifier rejects unsafe ones, and a motion generator executes the accepted skill in a half-physics simulator.
Related event: Meta's HumanCLAW Benchmark Shows Top VLMs Lack Body Awareness(3 posts)→
More from Embodied
- Transforming quadruped robot runs on real hardware: per-pose RL policies in Isaac switched on morph — _Stocko_ · 2026-10-16
- Foldable iPhone Duo reportedly sells out launch-day inventory, some deliveries slip to 2027 — Polymarket · 2026-10-16
- KinoGen synthesizes robot loco-manipulation reference motions directly via kinematically grounded video generation — ChongZzZhang · 2026-10-16
- Science Robotics paper introduces F³, predictive event-based vision beyond optical flow — CSProfKGD · 2026-10-16
- iPhone Duo foldable appears on South African retailer iStore at R47,999 in apparent leak — shauntrennery · 2026-10-16
- New paper: game-theoretic motion planning shows it can be legal to break the rules — GioeleZardini · 2026-10-16