APPLICATION OF MACHINE LEARNING METHODS TO A WHEELED MOBILE ROBOT TRAJECTORY TRACKING CONTROL

Abstract

This paper presents new results of an experimental study of the application of machine learning methods to trajectory tracking control of a wheeled mobile robot (WMR). The experiments were conducted using a custom-assembled wheeled mobile platform and software developed for its programmable components, enabling the implementation of WMR trajectory tracking control using machine learning. A model of a skid-steered wheeled mobile robot control was developed based on a recurrent neural network. The neural network was trained using experimental data obtained during testing of the robot control system, and it was demonstrated that the control and data collection system of the experimental robot ensures high accuracy in following a given trajectory. Specifically, the model constructed using machine learning methods allows predicting the robot's movement with high accuracy over a time horizon of approximately 3-4 seconds at an average robot speed of 1.2 meters per second. The results obtained during the trajectory tracking control of the experimental onboard control system are promising for skid-steered wheeled mobile robot of any carrying capacity and provide progress in the current field of machine learning applications in robotics, automatic control, and inertial navigation.