Abstract: Addressing the safety issues of intelligent vehicles operating in mixed traffic flows, an improved Q-learning-based algorithm for safe braking of intelligent vehicles in multiple scenarios is proposed. Firstly, a three-vehicle model is established based on road conditions and vehicle parameters, with scenarios of braking, following, and lane changing being simulated separately. Secondly, linear programming is applied to the training data to ensure the possibility of safe braking for intelligent vehicles. Finally, a reward function is designed to guide the agent in controlling the middle vehicle to remain between the front and rear vehicles while ensuring safe braking. Additionally, the algorithm incorporates an interval-block method to handle continuous state space issues. A comparative test between the proposed algorithm and the traditional Q-learning algorithm in three common driving scenarios reveals that the proposed algorithm outperforms the traditional Q-learning algorithm in terms of safety and training efficiency, effectively controlling the middle vehicle to remain between the front and rear vehicles while ensuring safety in braking, following, and lane changing scenarios.

Key words: Keywords: intelligent vehicle, intelligent braking, Q-learning, linear programming

摘要： 针对智能车辆在混合交通流下行驶的安全问题，提出一种基于改进Q学习的智能车辆多场景安全刹车算法。首先根据路面情况与车辆参数建立三车模型，并分别模拟刹车、跟车与变道场景；其次对训练数据进行线性规划，确保智能车辆存在安全刹车可能；最后通过奖励函数设置，引导智能体在保证安全刹车的基础上尽量控制中车位于前后两车中间。另外结合区间分块方法，使算法能处理连续状态空间问题。将提出的算法与传统Q学习算法在三种常见行驶场景下对比测试，仿真结果表明提出的算法相较传统Q学习算法安全性更好、训练效率更高，在刹车、跟车与变道场景中均能在保证安全的前提下尽可能控制中车位于前后两车中间。

关键词: 智能汽车, 智能刹车, Q学习, 线性规划