Scholarlink

<p>Collision avoidance (CA) is a fundamental problem for safe and efficient autonomous driving. In this paper, we proposed a novel CA method by integrating a risk evaluation strategy into deep reinforcement learning (DRL). The method formulates the CA problem in autonomous driving as a minimum risk optimization problem (MROP). In order to solve this formulated MROP, a risk evaluation model was first constructed to estimate the specific risk probabilities of a given driving state. Then the risk evaluation result was introduced to the reward function of a DRL. The DRL module was properly designed to obtain the optimal CA policy. Finally, the proposed method was tested on a traffic scenario with dense obstacles in Carla environment. Experimental results show the effectiveness of the proposed approach. Moreover, the effect of some DRL parameter settings on learning CA strategies was investigated.</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

Collision Avoidance for Autonomous Driving by Integrating Risk Evaluation Into Deep Reinforcement Learning

  • 期刊论文

  • 发表日期：

    2026年1月4日

查看全文
  • Xinjian Fan,
  • Yong Wang,
  • Xiaobo Che

摘要

Collision avoidance (CA) is a fundamental problem for safe and efficient autonomous driving. In this paper, we proposed a novel CA method by integrating a risk evaluation strategy into deep reinforcement learning (DRL). The method formulates the CA problem in autonomous driving as a minimum risk optimization problem (MROP). In order to solve this formulated MROP, a risk evaluation model was first constructed to estimate the specific risk probabilities of a given driving state. Then the risk evaluation result was introduced to the reward function of a DRL. The DRL module was properly designed to obtain the optimal CA policy. Finally, the proposed method was tested on a traffic scenario with dense obstacles in Carla environment. Experimental results show the effectiveness of the proposed approach. Moreover, the effect of some DRL parameter settings on learning CA strategies was investigated.