Research map / Reinforcement learning / Visual and representation RL
Learning Diverse Bimanual Dexterous Manipulation Skills from Human Demonstrations
Bohan Zhou, Haoqi Yuan, Yuhui Fu, Zongqing Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the simulation alignment stage with approaching, lifting and bonus rewards (ablation) | BiDexHD-IPPO teacher · TACO, Train | m2 tracking rate (%): 17.52 -> 78.18 | Improved |
| use functional grasping center instead of geometric center (ablation) | BiDexHD-IPPO teacher · TACO, Train | m2 tracking rate (%): 66.39 -> 78.18 | Improved |
| use functional grasping center instead of geometric center (ablation) | BiDexHD-IPPO teacher · TACO, Test New | m2 tracking rate (%): 22.63 vs 21.34 | No clear effect |
| add success bonus reward (ablation) | BiDexHD-IPPO teacher · TACO, Train | m2 tracking rate (%): 66.65 -> 78.18 | Improved |
| add success bonus reward (ablation) | BiDexHD-IPPO teacher · TACO, Test New | m2 tracking rate (%): 17.52 -> 21.34 | Mixed |
| use independent PPO instead of centralized PPO (ablation) | BiDexHD teacher · TACO, Train | m2 tracking rate (%): 53.88 -> 78.18 | Improved |
| distill IPPO teachers with DAgger into a vision policy | BiDexHD-IPPO · TACO, Test New | m2 tracking rate (%): 21.34 -> 53.71 | Improved |
| distill IPPO teachers with DAgger into a vision policy | BiDexHD-IPPO · TACO, Train | m2 tracking rate (%): 78.18 -> 74.59 | Hurt |
| use DAgger from RL teachers instead of behavior cloning from retargeted demonstrations | vision-based student · TACO, all splits | m2 tracking rate (%): 00.00 -> 74.59 (Train) | Improved |
| add K future object positions as student input (ablation) | BiDexHD-IPPO+DAgger · TACO, Train | m2 tracking rate (%): K=0: 72.09, K=5: 74.59 | No clear effect |
| add K=5 future object positions as student input (ablation) | BiDexHD-IPPO+DAgger · TACO, unseen tasks | m2 tracking rate (%): average 3.1% gain over K=0 | Improved |
Closest papers by meaning
- ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR 2025 · similarity 0.91)
- RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation (ICLR 2025 · similarity 0.89)
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 2025 · similarity 0.88)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- HumanoidGen: Data Generation for Bimanual Dexterous Manipulation via LLM Reasoning (NeurIPS 2025 · similarity 0.88)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
