Research map / Vision-language and multimodal / Robot action and navigation
On Robustness of Vision-Language-Action Model against Multi-Modal Perturbations
Jianing Guo, Zhenhong Wu, Chang Tu, Yiyao Ma, Xiangqi Kong, Zhiqian Liu, Jiaming Ji, Shuning Zhang, Yuanpei Chen, Kai Chen, Qi Dou, Yaodong Yang and 4 more
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 RobustVLA input and output robust training | pi0 · LIBERO, 17 perturbations average | success rate (%): 62.6 -> 76.6 | Improved |
| add RobustVLA training on clean tasks | pi0 · LIBERO, no perturbation | success rate (%): 96.0 -> 95.5 | No clear effect |
| add BYOVLA visual robustness (VLM inpainting) | pi0 · LIBERO, non-visual perturbations | success rate (%): +0.0% in non-visual modalities; +4.0% avg visual | No clear effect |
| add input regularization to output-only robust training (ablation) | RobustVLA · LIBERO, 17 perturbations average | success rate (%): 64.8 -> 76.6 | Improved |
| add output regularization to input-only robust training (ablation) | RobustVLA · LIBERO, 17 perturbations average | success rate (%): 71.7 -> 76.6 | Improved |
| add UCB-based perturbation selection (ablation) | RobustVLA · LIBERO, 17 perturbations average | success rate (%): 69.3 -> 76.6 | Improved |
| use random-perturbation domain randomization instead of UCB adversarial selection (ablation) | pi0 fine-tuning · LIBERO, 17 perturbations average | success rate (%): 62.6 -> 61.8 | No clear effect |
| output regularization alone on observation perturbations (ablation) | RobustVLA w/o in · LIBERO, Dead Pixel | success rate (%): 20.8 -> 40.8 | Improved |
| use larger PGD action epsilon (0.06 vs 0.03) (ablation) | RobustVLA · LIBERO, action noises average | success rate (%): 53.3 -> 53.2 | No clear effect |
| apply RobustVLA with limited real-world demonstrations | pi0 · Real-world FR5, 25 demos | success rate (%): 37.5 -> 92.5 | Improved |
| apply RobustVLA with abundant demonstrations | pi0 · Real-world FR5, 100 demos | success rate (%): 65.0 -> 95.0 | Improved |
Closest papers by meaning
- Scaling Test-Time Robustness of Vision-Language Models via Self-Critical Inference Framework (CVPR 2026 · similarity 0.90)
- MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs (ICML 2026 · similarity 0.90)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.89)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- ActionUNet: Improving Robustness of VLA Models with Efficient Multi-scale Fine-tuning (NeurIPS 2026 · similarity 0.89)
- ETA: Evaluating Then Aligning Safety of Vision Language Models at Inference Time (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.
