Latest revision as of 19:30, 3 February 2021

Abstract

From the accumulation of past and repeated experiences, driving a vehicle for most people has become almost an
automatism. People do it without being really conscious of all the multiple tasks involved. When it comes to
autonomous driving, it is a great challenge to transform this acquired knowledge into machine learning techniques.
Progressively deep learning has become the best tool to use for autonomous driving vehicle since it is possible to
emulate the behavior of the human brain in a large number of intelligent vehicles applications. The most common
use of this type of techniques has been the implementation of Convolutional Neural Networks (CNNs) for
classification and identification of obstacles and pedestrians in the vehicle’s surroundings. CNNs are especially
dedicated to image analysis and, even though they have been succesfully used for classification and pattern
learning, it is possible to use them for regression. Therefore, with a CNN architecture, continuous data can be
predicted, like other classical neural networks. On the other hand, an accurate knowledge of vehicle odometry is
of vital importance in autonomous driving. When exact positioning by GPS is not possible, knowing the trajectory
and specific location of vehicle become fundamental for safety. While using the advantages of CNN, this paper
presents a deep learning application that estimates continuously the vehicle speed and yaw rate to realize the
reconstruction of the car’s odometry. Since CNNs are suited for training with imagery, a 3D LiDAR sensor has
been used for the recognition of the environment as well as reconstruction of data-images. The results indicate that
the network’s architecture is able to estimate the speed and yaw rate from the LiDAR’s data-images. These facts
can be used to support autonomous navigation.

Original document

The different versions of the original document can be found in:

• https://zenodo.org/record/1441032 under the license http://creativecommons.org/licenses/by-nc-nd/4.0/legalcode

• https://zenodo.org/record/1441032,

http://dx.doi.org/10.5281/zenodo.1441031 under the license http://creativecommons.org/licenses/by-nc-nd/4.0/legalcode

• https://zenodo.org/record/1441032,

http://dx.doi.org/10.5281/zenodo.1441032 under the license http://creativecommons.org/licenses/by-nc-nd/4.0/legalcode

DOIS: 10.5281/zenodo.1441032 10.5281/zenodo.1441031