Navigability Estimation for Autonomous Vehicle Using Machine Learning

Abstract

Autonomous navigation is one fundamental problem in robotics. Most of the algorithms and solutions developed in this area was designed to operate in highly structured environments such as corridors and rooms inside buildings. However, another topic of great interest of researchers working in the field is about navigation on outdoor, unstructured environments, more specifically the development of intelligent autonomous vehicles.Since the 80's came the idea of developing intelligent autonomous vehicles, which have various applications: reduction in the number of accidents on streets and highways, increasing the efficiency of traffic in big cities and mobility for the disabled and the elderly. Despite the attention it receives by several researchers in robotics, there are still many open questionsas well as various challenges, and therefore it is subject to extensive scientific research due to the complexity of the problem addressed.For a robot/vehicle navigate safely is essential to accurate detect obstacles and navigable paths. Moreover, the methods developed for this purpose must maintain its accuracy in adverse situations, hence the importance of not relying on assumptions that restrict the functionality of the method to a particular environment.This research plan aims to study and develop innovative methods to detect obstacles andnavigable paths. Specifically, given a route by a specialist (human driver), we intend to extractfeatures associated with the navigability of the ground through machine learning algorithms. These characteristics are then used to determine the navigability of other land. Finally, as a possible way to demonstrate its application, it is expected to integrate the method developed with a navigation system.