Controller Gain Tuning of a Nonholonomic Mobile Robot via Neural Network Predictor

摘要

It is very important to design systems for people with visual loss due to increasing accidents in recent years or people born blind to increase their integration into their social lives. For this reason, in this paper, it is aimed to design a robotic system and control approach that will enable blind people to move from one place to another. In this regard, it is desired that a mobile robot can go to a target point with the least error on a certain trajectory in a working environment known with its obstacles. In accordance with this purpose, an analytical kinematic controller is designed, and it is proposed to determine control parameters via artificial neural network. The relationship between control gain values and mobile robot trajectory tracking errors is taught to two different artificial neural networks (Multilayer Perceptron and Radial Basis Neural Network) using experimental data obtained with the mobile robot. Then numerous tests are performed with these networks to determine the gain values with the lowest error. The real-time tracking performances of the determined gain values are experimentally investigated and revealed for two different platform conditions (with and without obstacles) and two different target points (A and B).