LeWAMapache-2.0
LeWAM Evaluation PushT π0.5 Frozen
Real-robot evaluation rollouts of the π0.5 policy (with frozen VLM backbone) on a Rebot B601 7-DoF arm performing non-prehensile PushT tasks. Contains 30 policy rollouts scored by human operators with a 10% success rate.
Downloads48
Episodes30
Hours0.35
Why This Matters for Physical AI
This dataset provides real-world policy evaluation data on a physical robot arm performing non-prehensile manipulation, demonstrating the performance gap between learned policies and ground truth on actual hardware.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- Rebot B601
- Action Space
- joint_positions
- Environment
- lab
- Task Types
- manipulationpush
- Episodes
- 30
- Total Hours
- 0.35
- Data Format
- LeRobot
- Annotation Types
- reward_labels
- License
- apache-2.0
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack