Diffusion Policy

Visuomotor policy learning via action diffusion. · By Columbia / Stanford

Diffusion Policy represents robot actions as a conditional denoising diffusion process, achieving strong results on multimodal manipulation tasks and inspiring many follow-ups.