podolinsky

GR00T-N1.7 LIBERO-10 backbone features + K=10 action samples (normal + occluded)

Aligned rollouts of the NVIDIA GR00T-N1.7-LIBERO policy on LIBERO tasks in normal and occluded variants, with K=10 action samples per inference and per-token backbone features from the Cosmos-Reason2-2B VLM backbone. Designed for representation probing and sampling-based failure detection research.

Downloads8
Episodes500

Why This Matters for Physical AI

This dataset enables research into failure detection and robustness of generative policies by providing multiple action samples and backbone features that allow analysis of model uncertainty and occlusion-induced failures in manipulation tasks.

Technical Profile

Modalities
rgbproprioceptionlanguage
Robot Embodiments
Franka Panda
Action Space
end_effector_delta
Environment
simulationkitchenliving_roomstudy
Task Types
manipulationpick_and_placeobject_placement
Episodes
500
Data Format
npz
Annotation Types
language_instructionsreward_labelsaction_labels
Part of the LIBERO family

Access

Need custom rgb data?

Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets