基于深度强化学习的自动驾驶分段决策方法  

Segmentation decision-making method for autonomous driving based on deep reinforcement learning

在线阅读下载全文

作  者:王春淇 张明恒 周俊平 姚宝珍[1] 石佳伟 WANG Chunqi;ZHANG Mingheng;ZHOU Junping;YAO Baozhen;SHI Jiawei(School of Mechanical Engineering,Dalian University of Technology,Dalian 116024,China)

机构地区:[1]大连理工大学机械工程学院,辽宁大连116024

出  处:《大连理工大学学报》2025年第2期142-151,共10页Journal of Dalian University of Technology

基  金:国家自然科学基金资助项目(52272413)。

摘  要:基于道路分段方法,在不同路段采用差异化的车辆控制策略,以满足自动驾驶决策系统的性能要求.首先,鉴于道路特征的复杂变尺度特性,通过小波变换和Otsu算法将道路特征与变换系数进行映射并实现道路自适应分段.其次,为提高模型决策能力,基于深度强化学习(deep reinforcement learning,DRL)与奖励分解架构将自动驾驶分段决策任务分解为横向、纵向两个并联决策子任务,分别构建分段决策模型与奖励函数,设计一种动作掩蔽策略来提升模型训练速度.最后,通过一系列实验验证所提模型的有效性.实验结果表明,与传统的DRL算法相比,引入奖励分解架构与动作掩蔽策略的DRL算法不仅保证驾驶系统可靠决策,而且在通行效率、安全性等方面也有所提升.Based on the road segmentation method,differentiated vehicle control strategies are adopted in different road segments to meet the performance requirements of autonomous driving decision-making systems.Firstly,considering the complex and variable scale characteristics of road features,the road features are mapped to transform coefficients using wavelet transform and Otsu algorithm to achieve adaptive segmentation of the road.Secondly,in order to improve the decision-making ability of the model,the autonomous driving segmentation decision-making task is decomposed into two parallel decision-making subtasks,horizontal and vertical,based on deep reinforcement learning(DRL)and reward decomposition architecture.The segmentation decision-making model and reward function are constructed separately,and an action masking strategy is designed to improve the training speed of the model.Finally,the effectiveness of the proposed model is verified by a series of experiments.The experimental results show that compared with the traditional DRL algorithm,the DRL algorithm with the introduction of reward decomposition architecture and action masking strategy not only ensures reliable decision-making of the driving system,but also improves traffic efficiency,safety and other aspects.

关 键 词:自动驾驶 道路分段 小波分解 深度强化学习 

分 类 号:U46[机械工程—车辆工程]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象