基于DQN改进的自动驾驶行为决策方法
An Improved Autonomous Driving Behavior Decision-making Method based on DQN
修彩靖1
作者信息
摘要
随着自动驾驶技术的不断发展,行为决策作为其中的关键技术之一,受到了广泛关注.文章提出了一种基于深度强化学习中的DQN(Deep Q-Network)改进的自动驾驶行为决策方法.该方法通过引入优先经验回放和双重DQN技术,提高了算法的收敛速度和稳定性.同时,针对自动驾驶多交互环境的复杂性,设计了合理的状态空间和动作空间,并进行了充分的实验验证.实验结果表明,该方法能够有效地实现自动驾驶车辆在多交互场景交叉路口的行为决策,提高了决策的通过性和场景泛化性.
Abstract
With the continuous development of autonomous driving technology,intersection behavior decision-making,as one of the key technologies,has received widespread attention.This paper proposes an improved autonomous driving intersection behavior decision-making method based on Deep Q-Network(DQN)in deep reinforcement learning.The method improves the convergence speed and stability of the algorithm by introducing prioritized experience replay and double DQN techniques.Meanwhile,aiming at the complexity of the intersection environment,a reasonable state space and action space are designed,and sufficient experimental verification is carried out.The experimental results show that this method can effectively realize the behavior decision-making of autonomous vehicles at intersections,improving the accuracy and real-time performance of decision-making.
关键词
自动驾驶/行为决策/深度强化学习/DQNKey words
Autonomous Driving/Behavior Decision-making/Deep Reinforcement Learning/DQN引用本文复制引用
出版年
2024
NETL
NSTL
万方数据