podolinsky
GR00T-N1.7 LIBERO-X backbone features + K=10 action samples
Aligned rollouts of a GR00T-N1.7 fine-tune on LIBERO-X simulator with K=10 action chunks sampled per policy inference, providing backbone features and multiple action samples for failure detection and representation probing research.
Downloads23
Episodes2180
Why This Matters for Physical AI
This dataset enables research on failure detection and representation probing for vision-language-action models by providing multiple action samples from the same observation alongside backbone features from a state-of-the-art generalist robot model.
Technical Profile
- Modalities
- videotabularbackbone_features
- Robot Embodiments
- humanoid
- Action Space
- end_effector_delta
- Environment
- simulation
- Task Types
- manipulationplace_on_surfaceplace_in_containerplace_relative_to_landmarkstack_objectsarticulate_closearticulate_opentoggle_appliance_ontoggle_appliance_off
- Episodes
- 2180
- Data Format
- npz
Access
Need custom video data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack