摘要
地下停车场环境复杂多变,使得寻车规划缺乏可靠的环境基础,易导致生成的路径并非全局最优,影响寻车效率。故提出基于多智能体深度策略梯度的地下停车场智能寻车方法。通过RFID虚拟标签技术的室内定位算法生成目标车辆的定位坐标,通过密集部署的虚拟标签实现对车辆的高精度定位;利用Shi-Tomasi角点检测方法生成地下停车场的角点坐标,并采用K-means聚类算法获取角点信息对应的地下停车场路口信息,规划出概率道路图(Probabilistic Roadmap, PRM);基于多智能体深度策略梯度算法强化学习PRM。每个智能体代表一条寻车路径,通过深度神经网络评估路径特征,实现路径的全局优化和动态调整,根据奖励函数结果输出最佳寻车方案。实验表明:该方法寻车路径的平均长度和平均拐点数量的数值水平均较低,奖励函数收敛速度快,说明该方法的路径规划性能高。
The complex and ever-changing environment of underground parking lots results in a lack of reliable environmental foundations for vehicle search planning,which can easily lead to the generation of paths that are not globally optimal and affect the efficiency of vehicle search.Therefore,this study proposes an intelligent car search method for underground parking lots based on multi-agent deep strategy gradient.Generate the positioning coordinates of the target vehicle through indoor positioning algorithms using RFID virtual tag technology,and achieve high-precision positioning of the vehicle through densely deployed virtual tags.Using the Shi-Tomasi corner detection method to generate the corner coordinates of an underground parking lot,and using the K-means clustering algorithm to obtain the intersection information of the underground parking lot corresponding to the corner information,a Probabilistic Roadmap(PRM)is schemed.Reinforce learning PRM based on multi-agent deep policy gradient algorithm.Each intelligent agent represents a vehicle search path,and evaluates path features through deep neural networks to achieve global optimization and dynamic adjustment of the path.The optimal vehicle search solution is output based on the reward function results.The experiment shows that after applying this method,the numerical levels of the average length and average number of turning points of the vehicle seeking path are relatively low,and the convergence speed of the reward function is fast,indicating that the path planning performance of this method is high.
作者
李静
汪震
Li Jing;Wang Zhen(School of Information Engineering,Anhui Vocational College of Electronics&Information Technology,Bengbu,Anhui 233000,China;College of Automotive Engineering,Jilin University,Changchun,Jilin 130000,China)
出处
《黑龙江工业学院学报(综合版)》
2025年第11期100-104,共5页
Journal of Heilongjiang University of Technology(Comprehensive Edition)
基金
安徽电子信息职业技术学院2024年度质量工程项目(自然科学类科研项目)“基于BP-KNN与改进A*算法复杂场景的智能寻车研究”(项目编号:2024kyxmzk004)。
关键词
多智能体深度强化学习
地下停车场
智能寻车
策略梯度
路径规划
multi-agent deep reinforcement learning
basement parking
intelligent car searching
strategic gradient
path planning
