Research map / Vision-language and multimodal / Robot action and navigation
Self-Improving Vision-Language-Action Models with Data Generation via Residual RL
Wenli Xiao, Haotian Lin, Andy Peng, Haoru Xue, Tairan He, Zhengyi Luo, Yuqi Xie, Fengyuan Hu, Linxi Fan, Guanya Shi, Yuke Zhu
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 |
|---|---|---|---|
| fine-tune on PLD data in addition to SFT | pi0 · LIBERO average | success rate (%): 93.4 -> 97.2 | Improved |
| fine-tune on PLD data in addition to OFT | OpenVLA · LIBERO average | success rate (%): 91.8 -> 99.2 | Improved |
| fine-tune on PLD data in addition to OFT | OpenVLA · LIBERO-Object | success rate (%): 99.1 -> 99.1 | No clear effect |
| fine-tune on PLD data | Octo · SimplerEnv average | success rate (%): 71.8 -> 96.6 | Improved |
| train on PLD data instead of base policy rollout data | pi0 · LIBERO-90 (full coverage) | success rate: 0.488 -> 0.871 | Improved |
| train on PLD data instead of human data | pi0 · LIBERO-90 (full coverage) | success rate: 0.815 -> 0.871 | Improved |
| train on PLD data instead of human data | pi0 · LIBERO-90 (0.1 coverage) | success rate: 0.272 -> 0.314 | Improved |
| use PLD data instead of human data | pi0 · Real-world blue cube pick-up | success rate (%): 40.0 -> 93.3 | Improved |
| use PLD data instead of human data | pi0 · Real-world red cube pick-up | success rate (%): 33.3 -> 66.7 | Improved |
| use PLD data instead of human data | pi0 · Real-world peg insertion | success rate (%): 100.0 -> 100.0 | No clear effect |
| increase base policy probing horizon alpha (ablation) | PLD · LIBERO-90 tasks 0-9 | success rate | Mixed |
| scale PLD data from 50 to 500 trajectories | pi0 · LIBERO-90 few-shot from LIBERO-Goal | success rate | Improved |
Closest papers by meaning
- What Can RL Bring to VLA Generalization? An Empirical Study (NeurIPS 2025 · similarity 0.91)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.91)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- RePO-VLA: Recovery-Driven Policy Optimization for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.90)
- EfficientVLA: Training-Free Acceleration and Compression for Vision-Language-Action Models (NeurIPS 2025 · 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.
