Applying convolutional neural networks in mobile robots navigation

Abstract

Providing robots with autonomous locomotion capability in different environments is crucial to expanding the usefulness of robotic systems. However, working out a mathematical or computational model of a mobile robot that can deal with any of the possible situations in real environments can be a complex task. For this reason, deep neural networks have been researched to solve the problems of autonomous navigation, using images obtained directly from a camera as input of the networks, thus bypassing the difficulties of mathematical modeling. This project aims to develop experiments to evaluate the use of neural networks and convolutional neural networks, as an alternative to overcome the typical problems of guidance and speed control in a small-scale vehicle model. Among the aspects to be investigated and evaluated in this work are the performance of the algorithm and the processing requirements when considering a low cost embedded system platform. (AU)