A Robot Learning from Demonstration Method Based on Neural Network and Teleoperation

Industrial robots are widely employed in electronics, aerospace, machining, and other fields due to their flexibility, efficiency, and accuracy characteristics. However, traditional robots necessitate skilled professionals to accomplish intricate programming tasks of trajectory planning through teaching pendant or offline programming, which imposes high demands on the programming skills of users and significantly affects the work efficiency of the robot. This paper develops a Learning from demonstration method based on the neural network and teleoperation to solve this problem. The method establishes a neural network model that utilizes the input data of the master side of the teleoperation system and the error of the slave robot to predict and compensate for the mapping error in the teleoperation, and optimizes the robot's reproducing trajectory through the extreme learning machine. Besides, a teaching process can be performed by non-professionals, and the robot can reproduce the operation trajectory according to the collected trajectory data, which solves the problems of long-time cost and high operator proficiency in the traditional robot programming process. This paper builds a teaching system and conducts experimental verification based on Omega-7 equipment and the UR robot. The results show that the established teleoperating system can reproduce the mission trajectory through a single demonstration operation, and the taught trajectory is smoother in the reproduction process after the training of the extreme learning machine. In conclusion, this paper provides a trajectory-optimized method of teaching robots without traditional programming.