A Study on Optimal Data Bandwidth of Recurrent Neural Network–Based Dynamics Model for Robot Manipulators

Abstract

In this article, a recurrent neural network (RNN)-based learning method is propdosed for achieving the overall dynamic model of robot manipulators. Several sections, e.g., data acquisition, learning model, hidden layers, nodes, activation function, and data bandwidth, are designed to make the RNN-based learning method establish the overall dynamic model of the robot manipulators. The proposed method has a key point that the optimal data bandwidth can be obtained by the loss function and its derivative in the robot manipulators. Since the data bandwidth is set to be effective in learning process, it helps to provide high learning hit rate while significantly reducing time-consuming tasks caused by trial and errors in any robot manipulators. From these benefits, the proposed method offers a compact form and simplicity so that it can produce the convenience of practicing engineers in industrial fields. The effectiveness of the proposed one is verified through experiments with three scenarios, which is compared with that of the original data bandwidth in a real robot manipulator.