An embedded deep learning neural network control for a wheeled mobile robot

Abstract

This paper proposes an adaptive tracking scheme for a 4-wheeled skid steering mobile robot (4-WSSMR) using diagonal recurrent neural network controller based on the hybrid deep learning algorithm (DRNNC-HDLA). For the developed DRNNC-HDLA structure, the diagonal recurrent neural network is constructed, whose initial weights values are obtained through the hybrid deep learning algorithm. It is a combination of the restricted Boltzmann machine and the self-organized map of Kohonen. The network weights and learning rate for the proposed scheme are updated based on the Lyapunov stability criteria to achieve the controlled system stability. To show the robustness of the proposed algorithm, the results are compared to other existing algorithms. The proposed algorithm is practically implemented for controlling a 4-WSSMR to show the ability of the proposed algorithm to deal with real applications. The effectiveness of the proposed approach is validated through extensive real-world experiments involving uncertainties and disturbances, demonstrating its capability to achieve accurate and reliable trajectory tracking. This work advances the field by offering a reliable control solution for mobile robots operating under challenging conditions.