重庆理工大学学报2025，Vol.39Issue(17)：1-12,12.DOI:10.3969/j.issn.1674-8425(z).2025.09.001
基于模型深度强化学习的智能车控制策略研究
Research on intelligent vehicle control strategies based on models and deep reinforcement learning
摘要
Abstract
The varied traffic environments and constant interactions among traffic participants pose big challenges to decision-making and control of autonomous driving systems.Although the model-free reinforcement learning method can be directly applied in many complex scenarios without precise environment modeling,it focuses only on learning the mapping relationship between actions and rewards in specific environments and thus fails to conduct modeling work on the internal structure and laws of the environment.In some scenarios without no previous training,the model-free reinforcement learning method often proves unsafe and fragile.To improve the decision-making and control for intelligent vehicles on highways,this paper proposes a model-based reinforcement learning algorithm.Specifically,it employs the model constructed in the training stage to guide the Monte Carlo Tree Search(MCTS)algorithm to conduct heuristic search,formulating forward-looking strategies.Experiments demonstrate the model-based reinforcement learning algorithm possesses better generalization ability in unknown scenarios and out performs the model-free algorithm.Compared with the online planning method based on MCTS,the model-based algorithm improves the execution efficiency and achieves better globally optimal effect.关键词
自动驾驶/深度强化学习/蒙特卡洛树搜索/决策控制Key words
autonomous driving/deep reinforcement learning/MCTS/decision-making and control分类
交通工程引用本文复制引用
胡博,李珂,张苏男,岳岩,邓康..基于模型深度强化学习的智能车控制策略研究[J].重庆理工大学学报,2025,39(17):1-12,12.基金项目
国家自然科学基金项目(51905061) （51905061）
