NaviDriveVLM: Decoupling High-Level Reasoning and Motion Planning for Autonomous Driving
This work provides a more efficient and effective approach to VLM-based autonomous driving, benefiting researchers and developers working on end-to-end driving systems.
The paper addresses the trade-off in VLM-based autonomous driving between strong semantic understanding (large models) and efficient, precise control (small models). They propose NaviDriveVLM, a decoupled framework that separates high-level reasoning from motion planning, achieving superior end-to-end motion planning performance on the nuScenes benchmark compared to large VLM baselines.
Vision-language models (VLMs) have emerged as a promising direction for end-to-end autonomous driving (AD) by jointly modeling visual observations, driving context, and language-based reasoning. However, existing VLM-based systems face a trade-off between high-level reasoning and motion planning: large models offer strong semantic understanding but are costly to adapt for precise control, whereas small VLM models can be fine-tuned efficiently but often exhibit weaker reasoning. We propose NaviDriveVLM, a decoupled framework that separates reasoning from action generation using a large-scale Navigator and a lightweight trainable Driver. This design preserves reasoning ability, reduces training cost, and provides an explicit interpretable intermediate representation for downstream planning. Experiments on the nuScenes benchmark show that NaviDriveVLM outperforms large VLM baselines in end-to-end motion planning.