PhysBrain 1.5: Unified Physical Foundation Model for Robotics

PhysBrain 1.5 integrates vision-language understanding with action generation and future state prediction. It uses a common learning framework to jointly optimize language, end-effector motion, and visual targets via autoregressive next-token prediction trained on human interaction videos.