56. Diffusion policies for robot control

Generate robot action sequences with diffusion and other generative policy models. Train on multimodal trajectories and evaluate action consistency, latency, uncertainty, and recovery behavior.