EgoSteer2024apache-2.0
EgoSteer Real-World Bimanual Teleoperation Dataset
A real-robot dataset containing 54,454 teleoperated episodes (192 hours, 20.75M frames) of bimanual dexterous manipulation across 193 tasks on a RealMan dual-arm robot with dexterous hands and egocentric RGB-D cameras, with free-form English language instructions for every episode.
Downloads28
Episodes54,454
Hours192.1
Likes3
Why This Matters for Physical AI
This dataset bridges manipulation priors from egocentric human videos to real robot embodiments through large-scale bimanual teleoperation demonstrations across diverse dexterous manipulation tasks, advancing the learning of steerable manipulation policies from egocentric visual observations.
Technical Profile
- Modalities
- rgbdepthproprioceptionlanguage
- Robot Embodiments
- RealMan RM75-6FRuiyan RY-H2 dexterous hands
- Action Space
- joint_positions
- Environment
- labtabletop
- Task Types
- manipulationpick_and_placegraspingnon-prehensilereorientationbimanualcontact-rich
- Episodes
- 54,454
- Total Hours
- 192.1
- Data Format
- LeRobot
- Annotation Types
- language_instructionstask_labelsaction_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