Uncovering and Mitigating Positional Blind Spots in Vision-Language-Action Models

2026-08-02

AI Summary

This work addresses positional blind spots (PBS) in vision-language-action (VLA) models for robotic manipulation—regions in the workspace where failure rates sharply increase due to minor, task-irrelevant changes in object positions, even when instructions and scenes remain unchanged. The study systematically demonstrates the prevalence and spatial clustering of PBS across five state-of-the-art VLA models. To mitigate this issue without requiring internal model access, the authors propose a black-box, two-stage framework: first, high-risk regions are precisely localized via grid-based sampling and a one-sided likelihood ratio test; then, local demonstration data from these regions are collected to fine-tune the policy using LoRA. Experiments on two benchmarks show PBS failure rates as high as 0.58; the proposed method achieves a blind spot detection F1 score of 0.678—significantly outperforming baselines—and reduces overall failure rates by 40.00%–85.19% while preserving global performance stability.

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%.

Problem

• Positional Blind Spots
• Vision-Language-Action Models
• Robotic Manipulation
• Spatial Competence
• Failure Probability

Innovation

• Positional Blind Spots
• Vision-Language-Action Models
• Black-box Testing
• LoRA Fine-tuning
• Robotic Manipulation