Research map / Vision-language and multimodal / Robot action and navigation
RehearseVLA: Simulated Post-Training for VLAs with Physically-Consistent World Model
Junjin Xiao, Yandan Yang, Xinyuan Chang, Ronghan Chen, Feng Xiong, Mu Xu, Wei-Shi Zheng, Qing Zhang
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 world-model RL post-training (RehearseVLA) | OpenVLA-OFT · LIBERO average (5 demos/task) | success rate (%): 74.85 -> 79.6 | Improved |
| add RL post-training in world model | OpenVLA-OFT · LIBERO-Long (5 demos/task) | success rate (%): 57.0 -> 57.8 | No clear effect |
| add RL post-training in world model | OpenVLA-OFT · LIBERO-Object (5 demos/task) | success rate (%): 74.2 -> 86.6 | Improved |
| add extra exploration data to world simulator (ablation) | OpenVLA-OFT + RehearseVLA · LIBERO-Goal (5 demos/task) | success rate (%): 68.4 -> 79.8 | Improved |
| add trained reward head (vs prompted VLM) (ablation) | OpenVLA-OFT + RehearseVLA · LIBERO-Goal (5 demos/task) | success rate (%): 68.4 -> 68.8 | No clear effect |
| add both extra data and reward head (ablation) | OpenVLA-OFT + RehearseVLA · LIBERO-Goal (5 demos/task) | success rate (%): 68.4 -> 86.4 | Improved |
| add both extra data and reward head (ablation) | OpenVLA-OFT + RehearseVLA · LIBERO-Long (5 demos/task) | success rate (%): 42.2 -> 57.8 | Improved |
| add reflector-based termination | OpenVLA-OFT + RehearseVLA · LIBERO average, no GT termination | success rate (%): 63.05 -> 74.9 | Improved |
| use RehearseVLA instead of simulator-based RIPT-VLA | OpenVLA-OFT · LIBERO suites (5 demos/task) | success rate (%): Object 83.4 -> 86.6; Spatial 88.6 -> 87.6 | Mixed |
| apply RehearseVLA post-training in real world | OpenVLA-OFT · Real-world tasks | success rate (%): Clean table 20 -> 30; green toy 30 -> 50 | Improved |
Closest papers by meaning
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.93)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- Knowledge Insulating Vision-Language-Action Models: Train Fast, Run Fast, Generalize Better (NeurIPS 2025 · similarity 0.90)
- VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model (ICML 2026 · similarity 0.90)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
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.
