Research map / Vision-language and multimodal / Robot action and navigation
DextER: Language-driven Dexterous Grasp Generation with Embodied Reasoning
Junha Lee, Eunha Park, Minsu Cho
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 contact-based embodied reasoning before grasp tokens (ablation) | DextER · DexGYS val | Success rate (%): 62.37 -> 67.14 | Improved |
| add contact-based embodied reasoning (ablation) | DextER · DexGYS val | Q1: 0.66 -> 0.89 | Improved |
| add contact-based embodied reasoning (ablation) | DextER · DexGYS val | diversity omega_q: 13.77 -> 13.63 | No clear effect |
| use PartField encoder instead of Uni3D (ablation) | DextER · DexGYS val | Success rate (%): 59.07 -> 67.14 | Improved |
| use 256 action bins instead of 512 (ablation) | DextER · DexGYS val | P-FID: 0.26 -> 0.20 | Improved |
| use 256 position bins instead of 128 (ablation) | DextER · DexGYS val | Success rate (%): 67.35 -> 67.14 | No clear effect |
| add contact position dropout of 0.5 (ablation) | DextER · DexGYS val | Success rate (%): 65.68 -> 67.14 | Improved |
| always drop contact positions (pdrop=1.0) (ablation) | DextER · DexGYS val | Success rate (%): 67.14 -> 63.33 | Hurt |
| scale LLM backbone from Qwen2.5-0.5B to 1.5B (ablation) | DextER · DexGYS val | Success rate (%): 67.14 -> 67.55 | No clear effect |
| use SmolLM2-360M backbone instead of Qwen2.5-0.5B (ablation) | DextER · DexGYS val | Success rate (%): 67.14 -> 64.87 | Hurt |
| specify more partial contact links as steering input (ablation) | DextER · Dexonomy Seen Obj. & Grasp | Success rate (%): 10.40 (1 link) -> 21.35 (5 links) | Improved |
| steering with contact links on unseen grasp taxonomy (ablation) | DextER · Dexonomy Unseen Grasp | Success rate (%): 7.08 (1 link) -> 9.10 (5 links) | Mixed |
Closest papers by meaning
- Grasp as You Say: Language-guided Dexterous Grasp Generation (NeurIPS 2024 · similarity 0.92)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- 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)
- RAGG: Retrieval-Augmented Grasp Generation Model (AAAI 2025 · similarity 0.88)
- Grasp2Grasp: Vision-Based Dexterous Grasp Translation via Schrödinger Bridges (NeurIPS 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.
