Research map / 3D vision / 3D generation
Scalable Cross-embodiment Dexterous Grasping via Morphology-Prior Diffusion
Sihang Li, Zheming Zhou, Marcelino Almeida, Omid Alizadeh, Luca Carlone, Min Sun, Chen Feng, Cheng-Hao Kuo
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 |
|---|---|---|---|
| scale from 3 to 7 hands | SOMO · Grasp synthesis | avg. success: 92.37 -> 78.93 (-13.4 pp), baselines 87.53 -> 64.45 and 93.90 -> 63.72 | Mixed |
| add more data (extended dataset with XHand) | SOMO · 7 hands | avg. success: 78.93 -> 81.02 | Improved |
| add more data (extended dataset) | D(R,O) and T(R,O) · 7 hands | avg. success: 64.45 -> 63.91 and 63.72 -> 65.87 | No clear effect |
| use classifier-free morphology prior for held-out hand | SOMO vs D(R,O) · annotation-free Allegro | success rate: 93.30% vs. 43.20% | Improved |
| use classifier-free morphology prior for held-out hand | SOMO vs D(R,O) · annotation-free LeapHand | success rate: 32.40% vs. 2.80% | Improved |
| add variable-length tokens instead of fixed-topology graph (ablation) | SOMO · 4 supervised hands | avg. success: +0.96 pp | Improved |
| replace BPS link encoding with VAE (ablation) | SOMO · 4 supervised hands | avg. success: +2.25 pp (79.38 -> 81.63) | Improved |
| replace VQVAE object encoding with VAE (ablation) | SOMO · 4 supervised hands | avg. success: +1.12 pp (81.63 -> 82.75) | Improved |
| guidance scale 1.25 instead of 1.0 for annotation-free hand (ablation) | SOMO · Barrett (annotation-free) | success rate: 81.4 -> 83.1 | Improved |
| guidance scale 1.0 instead of 1.25 for supervised hands (ablation) | SOMO · Allegro/Shadow avg. | success rate: 93.6 -> 94.2 | Improved |
| raise guidance scale to 2.0 (ablation) | SOMO · Barrett and avg. of supervised hands | success rate: 83.1 -> 79.9 (Barrett), 94.2 -> 89.9 (avg.) | Hurt |
| use SOMO under 50% partial point clouds | SOMO · 6 hands | avg. success: 70.70 vs 55.85 and 55.43 | Improved |
Closest papers by meaning
- Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp Generation (NeurIPS 2025 · similarity 0.89)
- LatentHOI: On the Generalizable Hand Object Motion Generation with Latent Hand Diffusion. (CVPR 2025 · similarity 0.89)
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger Bridges (NeurIPS 2025 · similarity 0.89)
- GEARS: Local Geometry-aware Hand-object Interaction Synthesis (CVPR 2024 · similarity 0.88)
- G-HOP: Generative Hand-Object Prior for Interaction Reconstruction and Grasp Synthesis (CVPR 2024 · similarity 0.88)
- DexGrasp Anything: Towards Universal Robotic Dexterous Grasping with Physics Awareness (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.
