DME-Driver: Integrating Human Decision Logic and 3D Scene Perception in Autonomous Driving

• DOI: 10.48550/arxiv.2401.03641
• Publication Date: 2024-01-01
• Authors: Wencheng Han; Dongqian Guo; Cheng‐Zhong Xu; Jianbing Shen
• Keywords: Control logic

In the field of autonomous driving, two important features driving car systems are explainability decision logic and accuracy environmental perception. This paper introduces DME-Driver, a new system that enhances performance reliability system. DME-Driver utilizes powerful vision language model as decision-maker planning-oriented perception control signal generator. To ensure explainable reliable decisions, logical is constructed based on large model. follows employed by experienced human drivers makes decisions in similar manner. On other hand, generation accurate signals relies precise detailed perception, which where 3D scene models excel. Therefore, planning oriented It translates made into for self-driving cars. effectively train proposed model, dataset was created. encompasses diverse range driver behaviors their underlying motivations. By leveraging this dataset, our achieves high-precision through thinking process.