Glossary · AI & ML
Diffusion policy
In brief
A diffusion policy is a robot control policy that uses a diffusion model — the same technique behind image generators like Stable Diffusion — to generate action sequences. The model learns to denoise random action noise into purposeful trajectories conditioned on observations.
Diffusion policies became prominent in robotics around 2023 (Cheng Chi et al., "Diffusion Policy: Visuomotor Policy Learning via Action Diffusion") because they handle multimodal action distributions cleanly. A regression model collapses to the mean when the demonstration data has multiple equally valid behaviors; a diffusion model can capture the full distribution and sample from it.
For humanoid manipulation tasks where there are several reasonable ways to grasp an object or several valid motion plans, diffusion policies tend to outperform behavior cloning. They're slower per inference call than direct policies, which matters at high control rates but is usually solved with hierarchical control.
Related terms
See it in the wild
Browse robots and brands using these techniques
Glossary entries are upstream. The catalog is where the implementations live.