Dynamic Simplified Model and Autotuning of Feedback Gain for Directional Control Using a Neural Network for a Small Tunneling Robot.

Abstract

This paper describes a simplified dynamic model and autotuning of feedback gain for the directional control of a small tunneling robot. First, we constructed a dynamic model for the amount of directional correction and determined its parameters by the least squares method. Next, we used a neural network to automatically obtain four feedback gains for the directional control of both pitching and yawing. The inputs for the neural network are an initial deviation and an initial angular deviation for pitching and yawing. The outputs of the neural network are the feedback gains for angular deviations and deviations. The neural network learns from the deviations obtained in the simulations. The neural network, which can adapt to any initial deviations, was formed using plural initial deviations in learning. Moreover, this method can tune the optimum gain for any design line. These results establish the validity of the proposed method.