Mocap2Combat

explores motion-prior reinforcement learning: agents are first trained on mocap data to learn human-like body dynamics, then adapted via self-play RL for physics-driven combat.