Scholarlink

<p>This paper proposed a human-like car-following model for intelligent vehicles based on deep reinforcement learning (DRL). To enhance anthropomorphic performance, a concise yet effective reward function is devised, incorporating desired inter-vehicle distance. Concurrently, the model’s state set, action set, neural network structure, and environment update strategy are meticulously crafted. The proposed model is then trained and tested on car-following events selected from the NGSIM dataset based on predefined selection rules. Then, a novel evaluation method employing the root mean square error (RMSE) of time headway (THW), time to collision inverse (TTCi), and acceleration is introduced to assess the anthropomorphic performance of the car-following model. Subsequently, the model’s efficacy is compared with benchmarks including the intelligent driver model (IDM), artificial neural network (ANN) models, model predictive control (MPC) models, and another variant of DRL models. Simulation findings unequivocally demonstrate the superior anthropomorphic performance of the proposed car-following model.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Modeling Human-Like Car-Following Model for Intelligent Vehicles Based on Deep Reinforcement Learning

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Keyin Wang,
  • Yiyang Liu,
  • Jianhui Zhang

摘要

This paper proposed a human-like car-following model for intelligent vehicles based on deep reinforcement learning (DRL). To enhance anthropomorphic performance, a concise yet effective reward function is devised, incorporating desired inter-vehicle distance. Concurrently, the model’s state set, action set, neural network structure, and environment update strategy are meticulously crafted. The proposed model is then trained and tested on car-following events selected from the NGSIM dataset based on predefined selection rules. Then, a novel evaluation method employing the root mean square error (RMSE) of time headway (THW), time to collision inverse (TTCi), and acceleration is introduced to assess the anthropomorphic performance of the car-following model. Subsequently, the model’s efficacy is compared with benchmarks including the intelligent driver model (IDM), artificial neural network (ANN) models, model predictive control (MPC) models, and another variant of DRL models. Simulation findings unequivocally demonstrate the superior anthropomorphic performance of the proposed car-following model.