Uncovering and Mitigating Positional Blind Spots in Vision-Language-Action Models
2026-08-03 • Robotics
Robotics
AI summaryⓘ
The authors studied how robot models that combine vision, language, and action sometimes fail in specific spots due to changes in objects that shouldn't matter. They call these tricky areas 'Positional Blind Spots' (PBS), where the robot suddenly makes more mistakes. To find these spots, the authors divide the workspace into grids and test which areas have higher failure risks. Then, they fix the problem by retraining the robot using examples from these spots, which helps improve performance locally without hurting the rest of the workspace. Their method worked well across different models and reduced failures by up to 85%.
Vision-Language-Action (VLA) modelsRobotic manipulationPositional Blind Spots (PBS)Log-likelihood-ratio testWorkspace gridFine-tuningLoRA (Low-Rank Adaptation)Failure rateTargeted retrainingBenchmark evaluation
Authors
Dongdong An, Pengjie Zhao, Yihao Huang, Wenbing Tang, Ziming He, Jiayi Zhu, Jifeng Ning, Qin Zhao
Abstract
Recent Vision-Language-Action (VLA) models achieve promising performance in robotic manipulation, typically measured by success rates aggregated over predefined object configurations, an evaluation that implicitly assumes spatially uniform competence across the workspace. However, this assumption does not hold: even with the instruction and every other scene factor held fixed, merely relocating a task-irrelevant distractor can sharply raise the failure probability within localized, spatially coherent regions, which we term Positional Blind Spots (PBS). In this paper, we propose a two-stage black-box framework to uncover and mitigate PBS. During the uncovering stage, we grid the workspace and apply a one-sided log-likelihood-ratio test to localize PBS cells with significantly elevated risk. During the mitigation stage, we fine-tune the policy via LoRA on demonstrations collected from these PBS regions, improving competence there while largely preserving performance across the rest of the workspace. We evaluate our framework on five state-of-the-art VLA policies across two benchmarks, and find that PBS are pervasive and spatially concentrated in all of them, with failure rates up to 0.58. Our search strategy achieves an average F1-score of 0.678, outperforming random search and adaptive sampling baselines by 0.268 and 0.178, respectively. Guided by the discovered regions, targeted fine-tuning reduces the overall failure rate by 40.00%--85.19%.