Learning to Assist: Physics-Grounded Human-Human Control via Multi-Agent Reinforcement Learning

CVPR 2026