A Low-Rank Method for Vision Language Model Hallucination Mitigation in Autonomous Driving

TL;DR

This paper introduces a low-rank, self-contained method to detect and rank hallucination-free captions generated by vision language models in autonomous driving, improving accuracy and efficiency without external references.

Contribution

A novel low-rank decomposition approach for ranking VLM captions based solely on their internal embeddings, enhancing hallucination detection in autonomous driving.

Findings

Achieves 87% accuracy in identifying hallucination-free captions

Reduces inference time by 51-67% compared to debate methods

Strong correlation between residuals and human hallucination judgments

Abstract

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 novel self-contained low-rank approach to automatically rank multiple candidate captions generated by multiple VLMs based on their hallucination levels, 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 one with the smallest residual as the most hallucination-free. Experiments on the NuScenes dataset demonstrate that our approach achieves 87% selection accuracy in identifying hallucination-free captions, representing a 19% improvement over the unfiltered baseline and a 6-10% improvement over multi-agent debate method. The sorting produced by sparse error 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 debate approaches, making it practical for real-time autonomous driving applications.