Scholarlink

<p>On-ramp merging is a common traffic scenario and it is important for autonomous vehicles to collaboratively learn a strategy to complete the merging process. In this study, we propose a reinforcement learning-based method incorporating Nash equilibrium to achieve an interactive merging strategy. First, game theory is utilized to model the interaction between the ramp vehicle and the mainline vehicle; a non-zero-sum game with refined reward settings is developed to reflect the different objectives and influential factors under different joint actions. The concept of Nash-<i>Q </i>learning is adopted, in which the optimal <i>Q</i>-value is based on the reward when all agents play Nash equilibrium strategies from the next period onward. Then Deep <i>Q</i>-Network (DQN) is implemented in training. The proposed method is trained and evaluated using field observation data of Hanshin Expressway in Osaka, Japan, resulting in improved driving performance.</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

A Nash Equilibrium-Based Reinforcement Learning Method for Autonomous Vehicle Merging Strategy

  • 期刊论文

  • 发表日期：

    2026年4月28日

查看全文
  • Chunyi Wei,
  • Toshiyuki Yamamoto

摘要

On-ramp merging is a common traffic scenario and it is important for autonomous vehicles to collaboratively learn a strategy to complete the merging process. In this study, we propose a reinforcement learning-based method incorporating Nash equilibrium to achieve an interactive merging strategy. First, game theory is utilized to model the interaction between the ramp vehicle and the mainline vehicle; a non-zero-sum game with refined reward settings is developed to reflect the different objectives and influential factors under different joint actions. The concept of Nash-Q learning is adopted, in which the optimal Q-value is based on the reward when all agents play Nash equilibrium strategies from the next period onward. Then Deep Q-Network (DQN) is implemented in training. The proposed method is trained and evaluated using field observation data of Hanshin Expressway in Osaka, Japan, resulting in improved driving performance.