AutoDrive-R²: Incentivizing Reasoning and Self-Reflection Capacity for VLA Model in Autonomous Driving

Vision-Language-Action (VLA) models in autonomous driving systems have recently demonstrated transformative potential by integrating multimodal perception with decision-making capabilities. However, the interpretability and coherence of the decision process and the plausibility of action sequences remain largely underexplored. To address these issues, we propose AutoDrive-R², a novel VLA framework that enhances both reasoning and self-reflection capabilities of autonomous driving systems through chain-of-thought (CoT) processing and reinforcement learning (RL). Specifically, we first propose an innovative CoT dataset named nuScenesR²-6K for supervised fine-tuning, which effectively builds cognitive bridges between input information and output trajectories through a four-step logical chain with self-reflection for validation. Moreover, to maximize both reasoning and self-reflection during the RL stage, we further employ the Group Relative Policy Optimization algorithm within a physics-grounded reward framework that incorporates spatial alignment, vehicle dynamic, and temporal smoothness criteria to ensure reliable and realistic trajectory planning. Extensive evaluation results across both nuScenes and Waymo datasets demonstrates the state-of-theart performance and robust generalization capacity of our method. the traffic light is red, the vehicle in front is moving forward, and the pedestrian is currently walking across the crosswalk, the vehicle must stop to ensure safety. 2. Calculation: Apply the kinematic equations to predict the future positions of the vehicle. Since the vehicle is must stop, no need for calculation. 3. Logical Deductions: Although there is a pedestrian crossing, the vehicle must halt due to the red light, regardless of the pedestrian's movement. 4. Reflection: Ensure that the predicted trajectory is collision-free and adheres to the lane marking and the road rules.