AI autonomous driving

DriveVLM is an autonomous driving system that utilizes visual language models (VLMs) to enhance scene understanding and planning capabilities. The system improves understanding of complex and long-tail scenarios through a unique combination of reasoning modules, including scenario description, scenario analysis, and hierarchical planning. Furthermore, to address the limitations of VLMs in spatial reasoning and computational requirements, DriveVLM-Dual is proposed, a hybrid system that combines the advantages of DriveVLM with traditional autonomous driving processes. Experiments on nuScenes dataset and SUP-AD dataset show that DriveVLM and DriveVLM-Dual are very effective in handling complex and unpredictable driving conditions. Finally, DriveVLM-Dual was deployed on a production vehicle, verifying its effectiveness in real-world autonomous driving environments.