Meta's HumanCLAW Benchmark Shows Top VLMs Lack Body Awareness

Meta and academic partners introduced HumanCLAW, a new embodied AI benchmark decoupling decision-making from motor control. Tests revealed that current top-tier vision-language models struggle significantly with physical body awareness, achieving only a 16.8% success rate.

2026-07-30 ~ 2026-08-01 · 3 related posts

• HumanCLAW: Top VLMs Lack Embodied Self-Awareness, Score Only 16.8% — metaresearch · 2026-07-30
• Meta's HumanCLAW Benchmark: Best VLM Scores Only 16.8% on Embodied Tasks — liuziwei7 · 2026-07-30
• HumanCLAW by Meta Reveals SOTA VLMs Are Surprisingly Bad at Embodied Action — liuziwei7 · 2026-08-01