arrow-hf2026
RoboTwin VLA Framework Chunk30 Evaluation
Evaluation dataset containing results from five-task RoboTwin evaluation runs using the VLA framework with action chunk size 30, including per-episode videos, tick-level traces, and constraint results.
Downloads0
Episodes140
Why This Matters for Physical AI
This evaluation dataset benchmarks vision-language-action models on manipulation tasks, providing insights into grounding and object detection performance for robotic control in vision-based systems.
Technical Profile
- Modalities
- rgbaction_chunk
- Action Space
- action_chunk
- Environment
- simulation
- Task Types
- place_object_basketplace_bread_skilletmove_can_potmove_pillbottle_padplace_empty_cup
- Episodes
- 140
- Data Format
- JSONL
- Annotation Types
- action_labelsbounding_boxes
Access
Need custom rgb data?
Claru builds purpose-built datasets for simulation applications with dense human annotations and quality assurance.
Request a Sample Pack