记忆推理的放射源抓取机器人运动规划

Motion planning of radioactive source grasping robot based on memory reasoning

针对机器人抓取铅罐内部放射源时,因铅罐的半封闭和强辐射环境导致机器视觉难以应用于放射源抓取的问题,提出一种记忆推理的强化学习抓取方法.基于机器视觉构建智能机器人抓取系统运动学模型,采用力觉反馈实现智能机器人与铅罐内部环境的交互,通过对历史抓取数据的记忆推理决策,实现对放射源的自主抓取.在机器人操作系统中使用Gazebo仿真器,分别采用蒙特卡罗采样法和基于记忆推理的强化学习抓取方法进行仿真.结果表明,基于记忆推理的强化学习抓取方法的平均抓取效率比蒙特卡罗采样法高84.67%,能高效地解决铅罐内部放射源的自主抓取问题.

Aiming at the problem that machine vision is difficult to be applied to the radioactive source grasping due to the semi-closed and strong radiation environment of the lead can,we propose a memory reasoning based reinforcement learning grasping method.The kinematics model of intelligent robot grasping system is constructed based on machine vision.The interaction between the intelligent robot and internal environment of lead cans is realized by force feedback.Through the memory reasoning decision of historical grasping data,the autonomous grasping of radioactive sources is realized.Using the Gazebo simulator in robot operating system(ROS),the Monte Carlo sampling method and reinforcement learning grasping method based on memory reasoning are simulated,respectively.The results show that the reinforcement learning grasping method based on memory reasoning achieves the average grasping efficiency of 84.67%higher than that of Monte Carlo sampling method and thus demonstrate that the reinforcement learning grasping method can effectively solve the problem of autonomous grasping of radioactive sources in lead cans.