首页|期刊导航|汽车工程学报|基于探索参数自学习的无人矿卡路径规划方法

基于探索参数自学习的无人矿卡路径规划方法OA

A Path Planning Method for Autonomous Mining Trucks Based on Self-Learning of Exploratory Parameters

中文摘要英文摘要

为保障无人驾驶矿卡装载过程的安全性与高效性,规划一条满足装载点位置与朝向角约束的全局路径至关重要.针对传统路径规划算法因采用固定探索参数效率低下,算法耗时较长,而基于学习的方法受限于训练场景有限,难以适应矿山复杂环境的问题,提出一种基于深度强化学习优化的混合A*路径规划方法(Deep Q Network-Hybrid A*,DQN-HA*).基于深度Q网络(DQN)设计了强化学习模型.构建了一种利用DQN优化混合A*探索参数的自适应规划框架,提高路径规划效率,并确保算法泛化能力.基于矿山场景数据对算法进行了测试验证,结果表明,与传统算法及基于学习的算法相比,所提DQN‑HA*在计算效率与路径质量上均具有更优表现.

To ensure the safety and efficiency of the loading process for autonomous mining trucks,it is essential to plan a global path that satisfies the position and heading constraints of the loading point.Traditional path planning methods are inefficient due to fixed exploration parameters,while learning-based methods have difficulty adapting to complex mining environments because of limited training scenarios.To address these challenges,this paper proposes a Hybrid A* path planning method optimized using deep reinforcement learning,referred to as Deep Q Network-Hybrid A*(DQN-HA*).The proposed method integrates a DQN-based reinforcement learning framework with the Hybrid A* algorithm to adaptively optimize the exploration parameters,thereby improving both planning efficiency and generalization capability.Finally,the proposed algorithm was tested and verified using data from mining scenarios.The results show that,compared with traditional and learning-based algorithms,the proposed DQN-HA*significantly improves the computational efficiency and optimizes the path quality.

梁紫涵;王亚飞;章翼辰;秦涛;田磊;王学鹏

上海交通大学,上海 200240上海交通大学,上海 200240上海交通大学,上海 200240中国重型汽车集团有限公司,济南 250101中国重型汽车集团有限公司,济南 250101潍柴动力股份有限公司,山东,潍坊 261061

信息技术与安全科学

自动驾驶路径规划深度强化学习混合A*

autonomous drivingpath planningdeep reinforcement learninghybrid A*

《汽车工程学报》 2026 (5)

758-769,12

国家自然科学基金项目(52372417)

10.3969/j.issn.2095‒1469.2026.05.04

评论