mahgoobi
sim-branching-data
A dataset of branching search trajectories in simulation where multiple action chunks are proposed from states and their outcomes are recorded, including counterfactual rollouts of abandoned branches. Contains 1,354 nodes with state snapshots and images across 28 searches on 14 manipulation tasks.
Downloads0
Episodes28
Why This Matters for Physical AI
This dataset enables training of action selection and planning models by providing counterfactual outcomes from branching search, allowing systems to learn which actions would have succeeded even when not executed.
Technical Profile
- Modalities
- rgbproprioception
- Action Space
- action_chunks
- Environment
- simulation
- Task Types
- manipulationpick_and_placegrasping
- Episodes
- 28
- Data Format
- parquet
- Annotation Types
- action_labelsreward_labels
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack