Development of a multi-sensored autonomous ground vehicle

Abstract

In this study, we investigated the problem of constructing a small sized driveby- wire ground vehicle with sensors and the problem of its autonomous navigation in unstructured outdoor environments. Although different sensor installations were tried on the vehicle, the focus was on the camera that would imitate the human perception using monocular cues in the acquired images. A high importance was given to a depth perception algorithm that is used to estimate the relative maximum distances of the obstacles around. A study on the importance of the features extracted from the images was also included so that the running time of the algoritm could be decreased, which would enable the system run in real world. We have seen that lesser number of features could well be used to estimate the depths in an image. The experiments were done on a rough terrain with trees and bushes around where the robot also had to cope with the physical conditions while trying to find the best direction to follow. The results were encouraging; the vehicle travelled longer distances; and even traversed the whole test area as the used methods got more developed, however, when the system is considered as a whole, we made an inference that more feasible models are required to compensate for erroneous estimations and, in the end, to make the navigation safer, since the final decision of the vehicle depends on many sequential parameters.