Donghyun12282026
LIBERO rollouts of the masked-external-camera demo pre-adapt pi0.5 (shifted views)
Rollouts of the pi0.5-LIBERO demo pre-adapt policy with external camera masked at inference, evaluated across three camera views (small, medium, large) in LIBERO-large tasks. The dataset contains both successful trajectories and all rollouts (success and failure) with step budgets and workspace constraints matching LIBERO evaluation.
Downloads0
Episodes800
Why This Matters for Physical AI
This dataset demonstrates how masked camera views and egocentric-only visual feedback improve robotic manipulation performance, providing insights into visual grounding and policy transfer for embodied AI systems.
Technical Profile
- Modalities
- rgbproprioception
- Action Space
- joint_actions
- Environment
- simulation
- Task Types
- manipulationgraspingpick_and_place
- Episodes
- 800
- Data Format
- LeRobot
- Annotation Types
- language_instructionsaction_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