jellyhoapache-2.0

BiGym, state observations, OGBench layout, uniform 20 Hz

BiGym task demonstrations replayed once and stored in OGBench's on-disk layout with state observations at uniform 20 Hz sampling rate. Includes 40 manipulation tasks with privileged object state information and per-episode success labels.

Downloads84
Episodes1655

Why This Matters for Physical AI

This dataset enables offline reinforcement learning research on complex manipulation tasks with realistic failure modes, providing both successful and failed trajectories to train robust robot learning policies.

Technical Profile

Modalities
proprioceptionobject_state
Robot Embodiments
Franka Panda
Action Space
joint_positions
Environment
simulation
Task Types
manipulationpick_and_placegraspingopeningclosingstackingreaching
Episodes
1655
Data Format
OGBench
Annotation Types
reward_labelsepisode_successterminals
License
apache-2.0
Part of the BiGym family

Access

Need custom proprioception data?

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

Request a Sample Pack

Related Datasets