基于元强化学习的电力巡检机器人自主越障控制研究

Research on Autonomous Obstacle Crossing Control of Electric Power Inspection Robot Based on Meta-reinforcement Learning

常规的电力巡检机器人自主越障控制方法以静态越障控制为主,无法自主识别前方障碍,出现越障控制失误的问题。因此,设计了基于元强化学习的电力巡检机器人自动越障控制方法。提取机器人自主越障动力学特征,将电力巡检机器人越障过程中受到的吸附力、支持力、摩擦力考虑在内,分析越障控制的静平衡条件,从而避免越障倾覆的问题。基于元强化学习构建巡检机器人自主越障控制模型,利用元强化学习算法自动学习越障控制模型的超参数,优化自主越障控制网络结构,实现机器人的精准控制。规划电力巡检机器人自主越障控制轨迹,在巡检机器人满足重力平衡条件的基础上,规划电力巡检机器人自主越障轨迹,通过机器人关节变化状态,达到控制机器人越障的目的。采用对比实验,验证了该方法的越障控制性能更佳,能够应用于实际生活中。

The conventional autonomous obstacle control method of electric inspection robot is mainly static obstacle control,which cannot independently identify the obstacles in front,and the problem of obstacle control error occurs.Therefore,the automatic obstacle crossing control method of electric power inspection robot based on meta-reinforcement learning is designed.The dynamic characteristics of the robot autonomous obstacle crossing are extracted,the adsorption force,support force and friction force received in the process of the power inspection robot are taken into account,and the static balance conditions of the obstacle crossing control are analyzed,so as to avoid the problem of obstacle crossing and overturning.Based on meta-reinforcement learning,the inspection robot autonomous barrier-crossing control model is constructed to automatically learn the super parameters of the meta-reinforcement learning algorithm to optimize the network structure of autonomous barrier-crossing control and realize the accurate control of the robot.Planning the trajectory of the autonomous obstacle control of the electric inspection robot.On the basis of the inspection robot meeting the gravity balance conditions,the autonomous obstacle control trajectory of the electric inspection robot is planned to control the obstacle control of the robot through the change state of the robot joint.By using comparative experiment,the method has better obstacle control performance and can be applied in real life.