Diffusion Policy: Visuomotor Policy Learning via…

Diffusion Policy asks how a robot can imitate demonstrations when several incompatible action sequences are all valid. Read it to see how denoising samples one coherent action chunk, why the controller predicts farther than it executes, how that receding horizon corrects drift, and what the reported 46.9% relative improvement, real-robot results, latency, and evaluation protocol actually establish.

Reading focus: Why averaging demonstrations can create an action that belongs to no successful strategy, while diffusion can sample one mode at a time. How observation, prediction, and execution horizons turn a denoised action sequence into a feedback controller. How to read the 15-task evaluation, 46.9% relative improvement, real Push-T result, and unresolved inference-step discrepancy without overstating them.

The International Journal of Robotics Research, 2024. Cheng Chi et al.. 35 min read, medium difficulty.