Research map / 3D vision / Human pose estimation
UniDexGrasp: Universal Robotic Dexterous Grasping via Learning Diverse Proposal Generation and Goal-Conditioned Policy
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a teacher policy with state canonicalization, object curriculum, and classification joint training (full method) vs. ILAD baseline | goal-conditioned state-based grasp policy · train set, state-based | success rate: 0.74 vs 0.25 | Helped |
| Remove state canonicalization (ablation) | full state-based teacher policy · train set, state-based | success rate: 0.74 -> 0.59 | Hurt |
| Remove object classification joint training of PointNet (ablation) | full state-based teacher policy · train set, state-based | success rate: 0.74 -> 0.65 | Hurt |
| Remove object curriculum learning (train directly on all categories) (ablation) | full state-based teacher policy · train set, state-based | success rate: 0.74 -> 0.31 | Hurt |
| Increase curriculum from 1-stage to 3-stage object curriculum (ablation) | teacher policy · train set, state-based | success rate: 1-stage 0.58, 2-stage 0.68, 3-stage 0.74 | Helped |
| Distill teacher into vision-based student policy | full pipeline · train set vs test set | success rate: drops by 6% (train) and 7% (test) | Hurt |
| Replace GraspGlow with CVAE (from GraspTTA) (ablation) | grasp proposal generation · seen cat, Q1 | Q1 and diversity: Q1 0.0423 -> 0.0362; CVAE collapses to a single mode | Hurt |
| Replace GraspIPDF rotation model with ReLie (ablation) | grasp proposal generation · seen cat, Q1 | Q1: 0.0423 -> 0.0190 | Hurt |
| Replace GraspIPDF rotation model with ProHMR (ablation) | grasp proposal generation · seen cat, Q1 | Q1: 0.0423 -> 0.0210 | Hurt |
| Remove test-time adaptation (TTA) (ablation) | grasp proposal generation · seen cat, Q1 and penetration | Q1 and object penetration: Q1 0.0423 -> 0.0013; penetration 0.205 -> 0.744 | Hurt |
| Decouple translation and joint angles into separate models (ablation) | grasp proposal generation · seen cat, Q1 | Q1: 0.0423 -> 0.0003 | Hurt |
Closest papers by meaning
- DexGrasp Anything: Towards Universal Robotic Dexterous Grasping with Physics Awareness (CVPR 2025 · similarity 0.91)
- UniGraspTransformer: Simplified Policy Distillation for Scalable Dexterous Robotic Grasping (CVPR 2025 · similarity 0.90)
- CADGrasp: Learning Contact and Collision Aware General Dexterous Grasping in Cluttered Scenes (NeurIPS 2025 · similarity 0.90)
- Cross-Embodiment Dexterous Grasping with Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger Bridges (NeurIPS 2025 · similarity 0.89)
- Generalizing 6-DoF Grasp Detection via Domain Prior Knowledge (CVPR 2024 · similarity 0.89)
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.