Researchers discovered that just three specific attention heads in frozen Vision-Language-Action (VLA) models can detect trajectory deviations with 44.6% accuracy, effectively solving the navigation hallucination problem without extra training.
arXiv · March 17, 2026 · 2603.13782
The Takeaway
This challenges the assumption that external critic modules or complex uncertainty heuristics are needed to detect VLA failures. It provides practitioners with a zero-cost, training-free way to monitor robot safety and trigger recovery policies in real-time.
From the abstract
Vision-Language-Action (VLA) models have demonstrated strong potential for predicting semantic actions in navigation tasks, demonstrating the ability to reason over complex linguistic instructions and visual contexts. However, they are fundamentally hindered by visual-reasoning hallucinations that lead to trajectory deviations. Addressing this issue has conventionally required training external critic modules or relying on complex uncertainty heuristics. In this work, we discover that monitoring