Digital Twins: A 3D simulation approach to test validation with an autonomous vehicle

Abstract

Within the Digital Twins context, efforts are being made to minimize costs, increase safety, and speed up tests within a specific application, based on computer graphics tools for three- dimensional simulations. Thus, as a way to mitigate mainly the risks with the safety issue involving activities with autonomous vehicles, the present work proposes the modeling and structuring of an electric golf cart in a 3D virtual environment, so that it can serve for studies in the areas of perception, navigation, and control. Therefore, taking as a reference the proper vehicle existing at the university, Blender software was used together with the Gazebo simulator to perform the validation of this simulation environment. Different open source Environment Mapping (SLAM), Pattern Recognition (YOLO), and Autonomous Navigation algorithms were integrated to minimize possible errors regarding their integration in the actual vehicle. Finally, validation tests were performed on the Yolo algorithm, resulting in an accuracy of 98.9% with a margin of error of 1.1% in the identification of objects.