Understanding Embodied Robotic Arms Through Diffusion Policy
0. Introduction This article introduces the “Diffusion Policy,” a new method for generating robot behaviors that represents the robot’s visuomotor policy as a conditional denoising diffusion process. It has been benchmarked across 15 different tasks in 4 different robot manipulation benchmarks, showing consistent superiority over existing state-of-the-art robot learning methods, with an average improvement of … Read more