DemoGrasp
BeingBeyond / DemoGrasp research team
ICLR 2026 research on learning dexterous grasping policies from a single demonstration and reinforcement learning.
What you can explore
Inspect the released demonstration replay, trained-policy evaluation and object-generalization settings. The quickstart uses an Inspire hand configuration.
Before you use it
Requires the documented Isaac Gym environment, external assets and checkpoints. Source instructions were reviewed; Dexhands has not reproduced the reported results. Physical transfer needs separate validation.
Original contributors
Haoqi Yuan, Ziye Huang, Ye Wang, Chuan Mao, Chaoyi Xu and Zongqing Lu
License / access: No reuse license verified in the repository; confirm code, asset and checkpoint permissions with the authors
Official code & reproduction instructions
Start with the right setup.
- Environment
- Isaac Gym Preview 4
- Embodiment
- Inspire quickstart configuration
What you need
The quickstart uses Python 3.8.19, Isaac Gym and external assets/checkpoints.
First useful check
Replay the documented demonstration, then evaluate the released policy.
Where it stops
Source instructions do not establish real-hand transfer. No code reuse license was verified.
Editorial reference · checked 2026-09-24