angelic123apache-2.0
RoboTwin evaluation: FastWAM stock vs. PhyPO
Evaluation outputs comparing FastWAM (Wan2.1 1.3B) stock model against PhyPO-trained variant on 18 of 50 RoboTwin dual-arm manipulation tasks, including logs, per-phase results, and episode videos under clean and randomized settings.
Downloads15
Episodes1800
Why This Matters for Physical AI
Provides comparative evaluation of vision language models on dual-arm robotic manipulation tasks under varying environmental conditions, demonstrating the impact of physics-aware training on model robustness and generalization.
Technical Profile
- Modalities
- rgb
- Robot Embodiments
- dual-arm
- Environment
- labsimulation
- Task Types
- manipulationpick_and_placegraspingopeningstackingpouringhandover
- Episodes
- 1800
- Data Format
- mp4
- Annotation Types
- language_instructionsreward_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