Research map / 3D vision / Human pose estimation
H-InDex: Visual Reinforcement Learning with Hand-Informed Representations for Dexterous Manipulation
Yanjie Ze, Yuyao Liu, Ruizhe Shi, Jiaxin Qin, Zhecheng Yuan, Jiashun Wang, Huazhe Xu
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 |
|---|---|---|---|
| use H-InDex representation instead of VC-1 | visual RL agent · 12 Adroit/DexMV tasks | normalized average score: 16.8% | Improved |
| use H-InDex representation instead of RRL (ImageNet ResNet-50) | visual RL agent · 12 Adroit/DexMV tasks | normalized average score: 25.4% | Improved |
| use H-InDex instead of VC-1 | visual RL agent · Pour | success rate: 67 -> 99 | Improved |
| use H-InDex instead of VC-1 | visual RL agent · Relocate tomato soup can | success rate: 30 -> 99 | Improved |
| use H-InDex instead of VC-1 | visual RL agent · Relocate mug | episode return | Hurt |
| use H-InDex instead of VC-1 on unseen backgrounds | visual RL agent · Relocate potted meat can, 9 novel backgrounds | score: 392.56 -> 475.04 | Improved |
| use H-InDex instead of VC-1 on unseen backgrounds | visual RL agent · Relocate potted meat can, 9 novel backgrounds | success rate: 16.8 -> 20.7 | Improved |
| use H-InDex instead of VC-1 on training background | visual RL agent · Relocate potted meat can, original scene | score: 2391.74 -> 2240.37 | No clear effect |
| add Stage 1 pre-training with 3D hand pose estimation (vs RRL, w/o Stage 1,2,3) (ablation) | H-InDex · 12 tasks | normalized score | Improved |
| add Stage 2 BatchNorm adaptation with keypoint detection (ablation) | H-InDex · 12 tasks | normalized score | Improved |
| add Stage 3 EMA BatchNorm in RL (ablation) | H-InDex · 12 tasks | normalized score | Improved |
| adapt 100% of parameters instead of 0.18% in Stage 2 (ablation) | H-InDex · 12 tasks | normalized score | Hurt |
Closest papers by meaning
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- Cross-Embodiment Dexterous Grasping with Reinforcement Learning (ICLR 2025 · similarity 0.90)
- ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR 2025 · similarity 0.89)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.88)
- HandEdit: A Unified Benchmark for Egocentric Human-to-Robot Dexterous Hand Image Editing (NeurIPS 2026 · similarity 0.88)
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 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.
