policy
SuSIE
UC Berkeley (RAIL Lab) · JAX
or hover any field below to flag it
Overview
Name
SuSIE
Author
UC Berkeley (RAIL Lab)
Framework
JAX
License
mit
Skill type
manipulation
Evidence level
verified
Task description
Hierarchical manipulation via subgoal synthesis with image-editing diffusion models. Finetunes InstructPix2Pix on robot video to generate visual subgoals from language commands. Low-level goal-conditioned diffusion policy executes toward each subgoal. Leverages Internet-scale visual pretraining for zero-shot generalization.
Spaces
Action space
end-effector-pose · 7-dim · 5Hz
Observation space
- type: multimodal
- · primary_rgb
- · language_instruction
- · generated_subgoal_image
Links
HuggingFace repo
Paper (arXiv)
Compatible robots
0+1 mentioned but not in catalog yetNo robots list SuSIE as compatible yet. Know of one? Flag it above.
Compatible environments
2Datasets that reference this policy
0No datasets reference SuSIE yet.