Research map / 3D vision / 3D generation
GraspXL: Generating Grasping Motions for Diverse Objects at Scale
Hui Zhang, Sammy Christen, Zicong Fan, Otmar Hilliges, Jie Song
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 objective-driven hand guidance (bias on wrist PD controller) (ablation) | GraspXL · ShapeNet test | Success rate (%): 68.5 -> 81.0 | Improved |
| add joint-to-surface distance features (ablation) | GraspXL · PartNet test | Success rate (%): 81.6 -> 95.0 | Improved |
| add two-phase learning curriculum (ablation) | GraspXL · PartNet test | Heading error (rad): 0.381 -> 0.270 | Improved |
| add two-phase learning curriculum (ablation) | GraspXL · PartNet test | Success rate (%): 96.2 -> 95.0 | No clear effect |
| train with 500 objects instead of 58 objects (ablation) | GraspXL · PartNet and ShapeNet test | Success rate (%): 95.0 -> 94.9 (PartNet), 81.0 -> 84.3 (ShapeNet) | No clear effect |
| replace reference-based SynH2R with RL policy without references | grasping motion synthesis · PartNet test | Success rate (%): 82.3 -> 95.0 | Improved |
| more motion objectives combined (direction only vs direction+rotation+midpoint) (ablation) | GraspXL · PartNet test | Success rate (%): 96.1 -> 95.0 | Hurt |
| randomize friction coefficient around default (ablation) | GraspXL · ShapeNet test | Success rate (%): 81.6 -> 80.5 | No clear effect |
| apply policy to different robot hands (Allegro, Shadow, Faive) | GraspXL · PartNet test | Success rate (%): 95.0 (MANO) vs 95.3, 94.0, 95.8 | No clear effect |
| apply policy to unseen ShapeNet objects instead of PartNet | GraspXL · ShapeNet vs PartNet | Success rate (%): 95.0 -> 81.0 | Hurt |
Closest papers by meaning
- UGG: Unified Generative Grasping (ECCV 2024 · similarity 0.91)
- Contact Map Transfer with Conditional Diffusion Model for Generalizable Dexterous Grasp Generation (NeurIPS 2025 · similarity 0.90)
- LatentHOI: On the Generalizable Hand Object Motion Generation with Latent Hand Diffusion. (CVPR 2025 · similarity 0.90)
- DexGrasp Anything: Towards Universal Robotic Dexterous Grasping with Physics Awareness (CVPR 2025 · similarity 0.89)
- RAGG: Retrieval-Augmented Grasp Generation Model (AAAI 2025 · similarity 0.89)
- GEARS: Local Geometry-aware Hand-object Interaction Synthesis (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.
