Autonomous navigation of mobile robot using shallow and deep neural network

Abstract

Neural network provides an efficient solution for the manoeuvrability problems of autonomous mobile robots. Freire et al. (2009) studied the autonomous navigation of SCITOS-G5 (mobile robot) by training four types of neural networks to perform a classification task. The multi-layer perceptron (MLP) network gave the best performance in comparison to others (mixture of experts, Elman and logistic perceptron). In this paper, a different model of MLP network has been trained using the same training set whose performance is found to be better than that obtained by Freire, in terms of faster computations and a higher success rate. Also, since deep learning started gaining popularity in terms of its association with the neural network therefore, a recurrent neural network model is developed whose performance is compared to a simple MLP network. The results show that it gives superior performance, in comparison to the developed MLP network, using fewer training samples.