SRD:Balancing Safety and Optimality based on RSS and Reinforcement Learning for Autonomous Driving

Deep reinforcement learning is currently regarded as one of the most effective approaches for addressing motion decision-making in autonomous driving within complex scenarios. Ensuring the safety of autonomous vehicles during the application of deep reinforcement learning has become a critical concern. This paper introduces a motion decision-making model based on safety reinforcement learning. To strike a balance between the safety and efficiency of driving strategies, a safety model incorporating RSS (Responsibility-Sensitive Safety) and lane-changing models is proposed. This model guides the autonomous driving system to explore optimal driving strategies with safety and efficiency as primary objectives, ensuring the practical safety of the trained deep reinforcement learning model during real-world applications. A series of comparative experiments is conducted in the highway-env simulation environment to demonstrate the performance of the proposed method. Experimental results indicate that the proposed model yields superior training outcomes and significantly enhances the safety of the deep reinforcement learning model throughout the training process.