부산대학교 도서관

While interpretable decision-making is pivotal in au-tonomous driving, research integrating natural language models remains a relatively untapped. To address this, we introduce a multi-modal instruction tuning dataset that facilitates language models in learning visual instructions across diverse driving scenarios. This dataset encompasses three primary tasks: conversation, detailed description, and complex reasoning. Capitalizing on this dataset, we present a multi-modal LLM driving assistant named VLAAD. After fine-tuned from our instruction-following dataset, VLAAD demonstrates proficient interpretive capabilities across a spectrum of driving situations. We open our work, dataset, and model, to public on github. https://github. com/sungyeonparkk/vision-assistant-for-driving