基于DDPG的爬塔机器人越障决策控制方法研究  

Research on Obstacle-Rossing Decision Control Method of Tower Climbing Robot Based on DDPG

作　　者：宁涵雪 张伟军[1] Ning Hanxue;Zhang Weijun(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《传动技术》2023年第4期7-13,共7页Drive System Technique

摘　　要：使用爬塔机器人完成高压输电铁塔的检修维护任务,要求机器人具有适应不同障碍环境的自动爬塔能力。不同铁塔的障碍大小、分布存在差异,传统预设逻辑的控制方法难以适应这种多样性,因此,提出了基于DDPG的越障控制方法。DDPG算法会根据机器人状态和环境感知系统获取周身障碍环境,进而探索不同机器人动作,并根据奖励信号调整确定最终的越障策略;同时基于机器人结构和任务需求设计行为树,将DDPG确定的策略映射到行为树节点上,实现具体的机器人动作。这种控制方法既能保证高层越障策略的灵活性,又允许机器人按照既定流程执行底层控制动作。通过实验证明了该方法有效提高了爬塔机器人的自主性和适应性。The completion of the maintenance and repair tasks on high-voltage transmission towers requires robots to have the ability to automatically climb the tower in different obstacle environments.The size and distribution of obstacles varies on different towers,which challenges the application of traditional preset logic control methods.To tackle this,a hierarchical control method is proposed based on the fusion of DDPG and behavior tree.The DDPG algorithm will firstly obtain the surrounding obstacle environment based on the robot's status and environmental perception system,explore different robot actions,and adjust the final obstacle clearance strategy based on reward signals.Based on the robot structure and task requirements,a behavior tree is subsequently designed to map the strategy determined by DDPG to the behavior tree nodes,achieving specific robot actions.This hierarchical control method not only ensures the flexibility of high-level obstacle crossing strategies,but also allows robots to execute low-level control actions according to established processes.The effective improvement in autonomy and adaptability of tower climbing robots using the proposed method has been proved by experiments.

关 键 词：强化学习 DDPG 越障 爬塔机器人 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]