nhatcm
HRI-diff large: HITBench closed-loop evaluation
A closed-loop evaluation dataset containing 180 object-perturbed tomato-to-bowl rollouts across nine human-robot interactions with 20 frozen states per interaction, including JSON summaries, trajectory arrays, and MP4 videos.
Downloads0
Episodes180
Why This Matters for Physical AI
This dataset provides real-world closed-loop evaluation data for assessing human-robot interaction policies under perturbations, critical for developing robust manipulation systems that can handle object variations and human collaboration.
Technical Profile
- Modalities
- rgbtrajectoryaction
- Action Space
- raw_policy_actions
- Environment
- lab
- Task Types
- pick_and_placehuman-robot-interaction
- Episodes
- 180
- Data Format
- JSONL
- Annotation Types
- action_labelstrajectory_labels
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack