Research map / Vision-language and multimodal / Robot action and navigation
Can VLMs Diagnose and Recover from VLA Manipulation Faults?
Bowen Yan, Jiahao Xiao, Kehui Liu, Jianbo Zhang, Zicheng Zhang, Qi Jia, Zhongjie Jia, Haoming Song, Chunyi Li, Bin Zhao, Guangtao Zhai
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 human-in-the-loop detect-rollback-correct feedback to the VLA (ablation) | OpenVLA-7B · LIBERO | average success rate: +13% points | Improved |
| add human-in-the-loop detect-rollback-correct feedback to the VLA (ablation) | Groot-N1.5-3B · Real robot | success rate: 30 -> 65 | Improved |
| roll back to a localized step instead of resetting to the initial pose (ablation) | human-corrected OpenVLA-7B · LIBERO | success rate: +8.3% points | Improved |
| add a GPT-5.2 VLM judger to the VLA | OpenVLA-7B · LIBERO-10 | simulation success rate: 43.00 -> 37.00 | Hurt |
| add a GPT-5.2 VLM judger to the VLA | OpenVLA-7B · LIBERO-Object | simulation success rate: 86.50 -> 70.50 | Hurt |
| add a Qwen3-VL-235B judger to the VLA | OpenVLA-7B · LIBERO-Object | simulation success rate: 86.50 -> 79.50 | Hurt |
| add a GPT-5-2 VLM judger to the VLA | Groot-N1.5-3B · Real robot | success rate: 30 -> 0 | Hurt |
| add a Qwen3-VL-235B judger to the VLA | Groot-N1.5-3B · Real robot | success rate: 30 -> 15 | Hurt |
| add a Gemini-2.5-Pro judger to the VLA | Groot-N1.5-3B · Real robot | success rate: 30 -> 20 | Hurt |
| correlate static fault-understanding ranking with simulation ranking | 20 VLMs · VLA-FixBench | rank correlation: Spearman 0.43, Kendall 0.33 | No clear effect |
Closest papers by meaning
- AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation (ICLR 2025 · similarity 0.88)
- SAFE: Multitask Failure Detection for Vision-Language-Action Models (NeurIPS 2025 · similarity 0.88)
- RoboProcessBench: Benchmarking Process-Aware Understanding in Visual-Language Robotic Manipulation (NeurIPS 2026 · similarity 0.86)
- Failing Forward: Adaptive Failure-Informed Learning for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.85)
- Code-as-Monitor: Constraint-aware Visual Programming for Reactive and Proactive Robotic Failure Detection (CVPR 2025 · similarity 0.85)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.85)
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.
