RNN-Enhanced Takagi-Sugeno Fuzzy Control for Tesla Model 3 Car-Like WMR Dynamics

Abstract

This paper investigates the integration of machine learning to enhance the performance of fuzzy control systems, specifically for steering control for Car-Like Wheeled Mobile Robot (WMR) dynamics. A Takagi-Sugeno fuzzy controller is employed to effectively manage the dynamic behavior of a Tesla Model 3, successfully controlling its steering across multiple trials. The non-linear vehicle model consistently follows the desired trajectories dictated by the fuzzy controller. To further enhance performance, data consisting of control signals and errors generated by the fuzzy controller are collected for various steering angles, and used to train a Recurrent Neural Network (RNN) to emulate and replace the controller. The trained RNN is subsequently tested on previously unseen steering angles, demonstrating a rapid response and accurate control signal generation. Remarkably, the RNN successfully generalizes beyond its training range, providing reliable control for steering angles outside the range of the training data. This study highlights the potential of hybrid fuzzy-Artificial Neural Network (ANN) systems to enhance the adaptability and efficiency of control strategies in autonomous vehicle applications.