点击次数：

论文类型：会议论文

发表时间：2019-08-08

收录刊物：EI

卷号：11745 LNAI

页面范围：368-378

摘要：In order to solve the problem of low-efficiency in the manual operation process of nameplate pasting for sand mold, an intelligent simulation system based on visual sensing and industrial robot is designed to paste nameplate on sand molds, and a deep reinforcement learning control method is proposed. The simulation system including the robot, visual sensor and sand mold is established in ROS combined with the physical simulation engine Gazebo. Then the task of nameplate pasting for sand molds is expressed as a markov process and the robot is trained by DQN method to learn a strategy to complete the task of pasting the nameplate of sand mold. A multi-level reward function algorithm based on multi-distances and collision information is proposed to improve the train success rate. Finally, the method is verified in the simulation system. The results show that the nameplate can be quickly attached to the sand mold cavity by the industrial robot. ? Springer Nature Switzerland AG 2019.