A low-rank consensus method for vision language model hallucination mitigation in autonomous driving

Vision Language Models (VLMs) are increasingly used in autonomous driving to help understand traffic scenes, but they sometimes produce hallucinations, which are false details not grounded in the visual input. Detecting and mitigating hallucinations is challenging when ground-truth references are unavailable and model internals are inaccessible. This paper proposes a self-contained low-rank approach to automatically rank multiple candidate captions generated by multiple VLMs according to their deviation from the shared semantic structure, using only the captions themselves without requiring external references or model access. By constructing a sentence-embedding matrix and decomposing it into a low-rank consensus component and a sparse residual, we use the residual magnitude to rank captions—selecting the candidate most strongly aligned with the estimated semantic consensus. Experiments on the NuScenes dataset demonstrate that our approach achieves 87% selection accuracy in identifying hallucination-free captions, representing a 16% improvement over the unfiltered baseline and a 5–8% improvement over multi-agent debate method. The sorting produced by sparse residual magnitudes shows strong correlation with human judgments of hallucinations, validating our scoring mechanism. Additionally, our method, which can be easily parallelized, reduces inference time by 51–67% compared to multi-agent debate methods, providing an efficient consensus-selection layer for end-to-end autonomous driving systems.