willi192026
AutoDex: Real-World Dexterous Grasp Experience
A dataset of 2,610 real-robot dexterous grasp trials collected autonomously with Allegro and Inspire hands, including multi-view video, calibrated camera parameters, 6-DoF object poses, and executed grasp annotations.
Downloads323
Episodes2610
Likes2
Why This Matters for Physical AI
AutoDex demonstrates autonomous data collection for dexterous manipulation, creating reusable grasp knowledge that generalizes across object instances through real-world trial-and-error learning.
Technical Profile
- Modalities
- rgbcamera_intrinsicscamera_extrinsicsproprioception
- Robot Embodiments
- xArm6 + AllegroxArm6 + Inspire
- Action Space
- joint_positions
- Environment
- lab
- Task Types
- dexterous_manipulationgrasping
- Episodes
- 2610
- Data Format
- numpy
- Annotation Types
- grasp_success_labelsaction_labelsobject_poseswrist_poses
Access
Need custom rgb data?
Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.
Request a Sample Pack