Recurrent Neural Network-Based Inverse Model Learning Control of Manipulators

This paper presents an inverse model learning trajectory control system of manipulators based on a second order recurrent neural network. The recurrent neural network approximates the inverse dynamic model of manipulators with less input information and simpler structure than the conventional applied feed-forward neural network. Based on analyzing the model of manipulators, the network structure and the learning algorithm are designed. Simulation experiments are carried out to demonstrate the performance difference between the system based on the recurrent neural network and that based on the feed-forward neural network. The results show that the former system has better performance in the model approximation efficiency, the control signal smoothness and the system robustness.