Robot Arm Task Imitation System Based on RNN

To simplify the complex motion planning problem of robot arm and make the robot arm have the generalization ability to adapt to the new task, a robot arm task imitation system based on recurrent neural network(RNN) was researched and implemented in this paper. First, the original task was taught by the teacher, and the teaching data was collected. Second, by constructing RNN to train the original teaching data, the control strategy of the robot arm imitation was obtained. Then, when the task changes, the movement of the new task was observed and the movement information was collected. Finally, the motion information was generalized via the robot arm control strategy based on RNN, and the control information of robot arm of imitating the new task was obtained, completing the imitation. Experimental results show that the method can obtain strategy simply and efficiently and has preferable generalization ability, which make the robot arm not only can imitate the original task, but also can imitate the new task when the task changes.