arXiv 2303.04137
Diffusion Policy: Visuomotor Policy Learning via Action Diffusion
By Cheng Chi, Zhenjia Xu, et al.
Published 2023-03-07
Citation lineage
Review the prior work and downstream research connected to this paper.
This paper introduces Diffusion Policy, a new way of generating robot behavior by representing a robot's visuomotor policy as a conditional denoising diffusion process. We benchmark Diffusion Policy across 12 different tasks from 4 different robot manipulation benchmarks and find that it consistently outperforms existing state-of-the-art robot learning methods with an average improvement of 46.9%. Diffusion Policy l…