Research map / Vision-language and multimodal / Robot action and navigation
DexVLG: Dexterous Vision-Language-Grasp Model at Scale
Jiawei He, Danshi Li, Xinqiang Yu, Zekun Qi, Wenyao Zhang, Jiayi Chen, Zhaoxiang Zhang, Zhizheng Zhang, Li Yi, He Wang
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 language instruction input (ablation) | DexVLG · LVIS-Seen | success rate: 84.4 -> 87.7 | Improved |
| add language instruction input (ablation) | DexVLG · Unseen | success rate: 64.5 -> 79.1 | Improved |
| add language instruction input (ablation) | DexVLG · SamPart3D | success rate: 70.2 -> 76.3 | Improved |
| add color to point cloud (ablation) | DexVLG · LVIS-Seen | success rate: 47.6 -> 75.3 | Improved |
| add color to point cloud (ablation) | DexVLG · LVIS-Seen | part grasp accuracy: 15.6 -> 39.1 | Improved |
| add color to point cloud (ablation) | DexVLG · SamPart3D | success rate: 33.6 -> 53.4 | Improved |
| scale training data from 10% to 100% (ablation) | DexVLG · LVIS-Seen | success rate: 49.7 -> 87.7 | Improved |
| scale training data from 10% to 100% (ablation) | DexVLG · Unseen | part grasp accuracy: 7.9 -> 36.6 | Improved |
| increase model size from 255M to 1B (ablation) | DexVLG · LVIS-Seen | success rate: 74.8 -> 75.3 | No clear effect |
| use 320M model instead of 255M or 1B (ablation) | DexVLG · LVIS-Seen | part grasp accuracy: 35.6 -> 41.6 | Mixed |
| add CLIP text embedding to small baseline | DGN2.0* · LVIS-Seen | success rate: 70.8 -> 67.8 | Hurt |
| add CLIP text embedding to small baseline | DGN2.0* · LVIS-Seen | part grasp accuracy: 38.5 -> 40.5 | Improved |
Closest papers by meaning
- RealVLG-R1: A Large-Scale Real-World Visual-Language Grounding Benchmark for Robotic Perception and Manipulation (CVPR 2026 · similarity 0.91)
- Grasp as You Say: Language-guided Dexterous Grasp Generation (NeurIPS 2024 · similarity 0.90)
- DexGrasp Anything: Towards Universal Robotic Dexterous Grasping with Physics Awareness (CVPR 2025 · similarity 0.89)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.89)
- Language-driven Grasp Detection (CVPR 2024 · similarity 0.88)
- DexFuncGrasp: A Robotic Dexterous Functional Grasp Dataset Constructed from a Cost-Effective Real-Simulation Annotation System (AAAI 2024 · 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.
