Treffer: 基于物理信息强化学习的无人驾驶车辆跟驰控制模型.

Car-following control is a fundamental technique for autonomous driving. In recent years, reinforcement learning has been widely adopted in car-following tasks, enabling models to exhibit strong learning and imitation capabilities. However, reinforcement learning-based models face challenges such as poor interpretability and unstable outputs, which pose potential safety risks. To address these issues, this paper proposed a physics-informed reinforcement learning car-following model. The model incorporated vehicle dynamics, defined continuous state and action spaces, and integrated three classical car-following models with reinforcement learning to enhance stability and interpretability. It constructed a simulation environment by using Python and the SUMO traffic simulator to train the PIRL-CF model. Comparative experiments were conducted against traditional car-following models and mainstream deep reinforcement learning models (DDPG and TD3). Experimental results show that the PIRL-CF model improves the proportion of comfort zones by 8% compared to deep reinforcement learning models. Additionally, it increases the minimum time-to-collision by 0.3s and the average headway distance by 0.21s compared to traditional models. These results demonstrate that the PIRL-CF model achieves a balance of safety, comfort, and dri-ving efficiency in car-following tasks, providing an effective solution for autonomous driving decision-making. [ABSTRACT FROM AUTHOR]

车辆跟驰控制是无人驾驶的基础控制技术之一。 近年来, 强化学习被广泛应用于无人驾驶车辆的跟驰 控制任务中, 使模型具备了较强的学习和模仿能力, 但也面临可解释性差和输出不稳定的问题, 给车辆运行带来 了潜在的安全隐患。 为了解决这些问题, 提出了一种融合强化学习与物理信息的车辆跟驰控制模型( physicsinformed reinforcement learning car-following model, PIRL-CF)。 该模型结合车辆动力学特性, 定义了连续的状态 集、动作集和奖励函数, 并引入三种经典物理跟驰模型与强化学习模型进行融合, 从而提升了模型的稳定性和可 解释性。 通过 Python 与交通仿真软件 SUMO 构建仿真测试平台, 对 PIRL-CF 模型进行了训练, 并与传统车辆跟 驰模型和主流深度强化学习模型 (DDPG 和 TD3) 进行了对比实验。 实验结果表明, 与深度强化学习模型相比, PIRL-CF 模型的乘车舒适区占比提高了 8%; 与传统物理跟驰模型相比, PIRL-CF 模型在最低碰撞时间上提升了 0.3 s, 平均车头时距提升了 0.21 s。 研究表明, PIRL-CF 模型能够在无人车跟驰控制任务中兼顾舒适性、安全性 和行车效率, 为无人驾驶智能决策提供了一种有效的技术方案。 [ABSTRACT FROM AUTHOR]

Copyright of Application Research of Computers / Jisuanji Yingyong Yanjiu is the property of Application Research of Computers Edition and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)