Diffusion Policy
Visuomotor policy learning via conditional diffusion
Diffusion Policy borrows the same technology behind AI image generators (like Stable Diffusion) and applies it to robot movement. Instead of generating a picture, it generates the robot's next sequence of moves — and it handles uncertainty and multiple possible actions far better than older approaches.
Well-documented with active community support. Mid-range GPU required. If you're comfortable with Python and basic ML, you can run demos in a few hours.
Learns robot visuomotor policies by representing the action distribution as a conditional denoising diffusion process. State-of-the-art across simulation and real-robot benchmarks.
Various manipulation demonstration datasets
Precise part insertion with variable starting positions
Gold standard baseline for manipulation papers
Pouring and placing tasks with natural variation
Requires some ML and Python experience
RELATED DISCUSSIONS
Community →Seeded and planned prompts — not live forum activity yet.
- Community signalIntermediate / AdvancedHow should EAR track open robotics models?
Meta-discussion on ontology, benchmarks, and graph links for open-weight VLAs and manipulation models.
- Community signalIntermediate / AdvancedOpenVLA in the field — what breaks first?
Open Models room prompt linked to openvla-7b — inference, fine-tuning, and sim-to-real gaps. Seeded prompt only.