HumanCLAW: Can Vision-Language Models Act Through a Body?

Meta introduces HumanCLAW-Bench, 1,218 long-horizon egocentric find-navigate-interact episodes across 41 indoor scenes, and a framework that separates decision-making from motor execution to isolate a VLM's 'action intelligence.' The best of nine tested state-of-the-art models achieved only 16.8% success, with failures traced to lost embodied self-awareness rather than poor target recognition.