Research map / Large language models / Alignment and preferences
EvolvingGrasp: Evolutionary Grasp Generation via Efficient Preference Alignment
Yufei Zhu, Yiming Zhong, Zemin Yang, Peishan Cong, Jingyi Yu, Xinge Zhu, Yuexin Ma
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 physical-constraint guidance in distillation (PGD) to the consistency model (ablation) | EvolvingGrasp CM · MultiDex (4-step) | Suc.6: 60.0 -> 64.3 | Improved |
| add physical-constraint guidance in sampling (PGS) (ablation) | EvolvingGrasp CM · MultiDex (4-step) | Suc.6: 60.0 -> 66.2 | Improved |
| add PGS on top of PGD (ablation) | EvolvingGrasp · MultiDex (4-step) | Suc.6: 64.3 -> 75.3 | Improved |
| add PGS to PGD, cost in time (ablation) | EvolvingGrasp · MultiDex (4-step) | Time (s): 0.10 -> 1.41 | Hurt |
| add HPO preference finetuning to full physics-aware model (ablation) | EvolvingGrasp · MultiDex (4-step) | Suc.6: 75.3 -> 76.8 | Improved |
| add HPO to CM+PGD without sampling guidance (ablation) | EvolvingGrasp · MultiDex (4-step) | Suc.6: 64.3 -> 67.5 | Improved |
| add HPO finetuning at 8 steps (ablation) | EvolvingGrasp · RealDex | Suc.6: 63.6 -> 64.4 | Improved |
| add HPO finetuning at 4 steps (ablation) | EvolvingGrasp · RealDex | Suc.6: 51.6 -> 50.6 | Hurt |
| add HPO finetuning in cross-dataset test (ablation) | EvolvingGrasp · RealDex (train DexGraspNet) | Suc.6: 52.6 -> 54.1 | Improved |
| increase sampling steps from 2 to 8 | EvolvingGrasp · MultiDex | Suc.6: 65.9 -> 80.3 | Improved |
| use real-time sampling without guidance (4-step) instead of guided | EvolvingGrasp · MultiDex | Suc.6: 76.8 -> 64.3 | Hurt |
| raise beta and N_ft at small step counts (ablation) | EvolvingGrasp HPO · Multidex (2-step) | Suc.6: 65.6 -> 65.9 | No clear effect |
Closest papers by meaning
- RAGrasp: A Retrieval-Augmented Framework with Diversity-Aware Modeling for Dexterous Grasp Generation (ECCV 2026 · similarity 0.89)
- Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp Generation (NeurIPS 2025 · similarity 0.89)
- ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR 2025 · similarity 0.88)
- UniGraspTransformer: Simplified Policy Distillation for Scalable Dexterous Robotic Grasping (CVPR 2025 · similarity 0.88)
- Cross-Embodiment Dexterous Grasping with Reinforcement Learning (ICLR 2025 · similarity 0.87)
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 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.
