Research map / Reinforcement learning / Visual and representation RL
DexNDM: Closing the Reality Gap for Dexterous In-Hand Rotation via Joint-Wise Neural Dynamics Model
Xueyi Liu, He Wang, Li Yi
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 joint-wise neural dynamics residual to the base policy (ablation) | generalist policy · Real, regular objects, palm down, z-axis | Rot (rad): 11.69 -> 23.82 | Improved |
| replace joint-wise dynamics model with whole-hand dynamics model (ablation) | DexNDM · Real, small objects, palm down, z-axis | Rot (rad): 9.29 -> 0.00 (z-axis, vs direct transfer 6.94) | Hurt |
| use joint-wise instead of whole-hand dynamics model (ablation) | DexNDM · Real, irregular objects, palm down, z-axis | Rot (rad): 0.72 -> 8.61 | Improved |
| joint-wise vs whole-hand dynamics with ample in-domain sim data (ablation) | dynamics model · in-domain, 3.1M sim trajectories | prediction accuracy | No clear effect |
| use joint-wise instead of whole-hand model under low data or OOD shift (ablation) | dynamics model · 7.5k real trajectories, OOD Thumb Up | prediction error | Improved |
| add UAN or ASAP style compensators trained on object-free data | base policy · Real world cylinder rotation | success | Hurt |
| add iterative real-world data collection rounds (ablation) | DexNDM · real data scaling | performance | No clear effect |
| add more real-world data (ablation) | DexNDM · real data scaling | performance | Improved |
| add DexNDM compensation in sim-to-sim transfer | base policy · Sim-to-sim MuJoCo | RotR: 82.03 -> 124.69 | Improved |
| ASAP compensator in sim-to-sim transfer | base policy · Sim-to-sim MuJoCo | RotR: 82.03 -> 26.25 | Hurt |
Closest papers by meaning
- World Models for Learning Dexterous Hand-Object Interactions from Human Videos (ECCV 2026 · similarity 0.92)
- ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR 2025 · similarity 0.90)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.89)
- DexTrack: Towards Generalizable Neural Tracking Control for Dexterous Manipulation from Human References (ICLR 2025 · similarity 0.88)
- DexHandDiff: Interaction-aware Diffusion Planning for Adaptive Dexterous Manipulation (CVPR 2025 · similarity 0.88)
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.
