Research map / Vision-language and multimodal / Robot action and navigation
Grounding Sim-to-Real Generalization in Dexterous Manipulation: An Empirical Study with Vision-Language-Action Models
Ruixing Jin, ZiCheng Zhu, Ruixiang Ouyang, Sheng Xu, Bo Yue, Zhizheng Wu, Guiliang Liu
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 table height randomization alone (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 2.7 -> 36.9 | Improved |
| add camera pose randomization alone (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 2.7 -> 23.5 | Improved |
| add background randomization alone (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 2.7 -> 11.5 | Improved |
| add table distractor randomization alone (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 2.7 -> 3.1 | No clear effect |
| combine all five randomization factors (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 2.7 -> 49.7 | Improved |
| add appearance randomization (BG) to spatial TH+CP (ablation) | SFT OpenVLA-OFT · Place Empty Cup (real) | success rate: 24.8 -> 33.5 | Improved |
| add distractor randomization to spatial TH+CP (ablation) | SFT OpenVLA-OFT · Place Empty Cup (real) | success rate: 24.8 -> 26.3 | Mixed |
| resample camera pose every frame instead of per episode (ablation) | SFT OpenVLA-OFT · Click Bell (real) | success rate: 23.5 -> 31.9 | Improved |
| resample background every frame instead of per episode (ablation) | SFT OpenVLA-OFT · Place Empty Cup (real) | success rate: 10.2 -> 23.1 | Improved |
| resample lighting every frame instead of per episode (ablation) | SFT OpenVLA-OFT · Place Empty Cup (real) | success rate: 8.6 -> 9.0 | No clear effect |
| raise rendering fidelity from low to medium (ray tracing on) (ablation) | All Factors DR policy · Place Empty Cup (real base) | successes / trials: 8/40 -> 12/40 | Improved |
| raise rendering fidelity from medium to high (ablation) | All Factors DR policy · Place Empty Cup (real base) | successes / trials: 12/40 -> 22/40 | Mixed |
Closest papers by meaning
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.91)
- GENMANIP: LLM-driven Simulation for Generalizable Instruction-Following Manipulation (CVPR 2025 · similarity 0.91)
- Scaffolding Dexterous Manipulation with Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- Robots Pre-train Robots: Manipulation-Centric Robotic Representation from Large-Scale Robot Datasets (ICLR 2025 · similarity 0.89)
- Position: Scaling Simulation is Neither Necessary Nor Sufficient for In-the-Wild Robot Manipulation (ICML 2024 · similarity 0.89)
- Predictive Inverse Dynamics Models are Scalable Learners for Robotic Manipulation (ICLR 2025 · 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.
