Our paper “AssistMimic: Learning to Assist — Physics-Grounded Human-Human Control via Multi-Agent Reinforcement Learning” has been accepted at CVPR 2026.

This work formulates the imitation of closely interacting, force-exchanging human-human motions as a multi-agent reinforcement learning problem. By jointly training partner-aware policies for both the assistant and the recipient in physics simulation, AssistMimic achieves physically grounded and socially meaningful control of assistive human-human interactions.

Authors: Yuto Shibata, Kashu Yamazaki, Lalit Jayanti, Yoshimitsu Aoki, Mariko Isogawa, Katerina Fragkiadaki (Carnegie Mellon University, Keio University, Keio AI Research Center)

Project: Coming soon