Physical Intelligence — Simar Kareer, Karl Pertsch, James Darpinian
Physical Intelligence reports that its vision-language-action model π 0.5 achieves roughly 2x performance gains on robotic tasks by learning from human video data at scale, demonstrating emergent transfer capabilities in robotic foundation models.