April-Yang2026
HumanIntent 六任务训练数据
A collection of seven frozen datasets used for training the pi0.5 model, containing real robot demonstrations and human replay data generated through three different pathways (AG/AH/OH), comprising 2,074 episodes across six manipulation tasks.
Downloads25
Episodes2074
Why This Matters for Physical AI
This dataset provides multi-modal real robot and human replay demonstrations for training embodied AI models on dual-arm manipulation tasks, enabling learning of complex object handover and manipulation skills.
Technical Profile
- Modalities
- rgbproprioception
- Robot Embodiments
- dual-arm manipulator
- Action Space
- end_effector_pose
- Environment
- lab
- Task Types
- handover_bottlepick_diverse_bottlesplace_bread_basketpnp_breadpnp_traystack_cups
- Episodes
- 2074
- Data Format
- LeRobot
- Annotation Types
- language_instructionstask_index
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack