Open AccessReal-time Driving Context Understanding using Deep Grid Net: A Granular Approach
Author(s) -
Liviu Marina
Publication year - 2020
Publication title -
international journal of robotic computing
Language(s) - English
Resource type - Journals
ISSN - 2641-9521
DOI - 10.35708/rc1869-126262
Subject(s) - context (archaeology) , occupancy grid mapping , computer science , hyperparameter , automotive industry , artificial intelligence , deep learning , grid , real time computing , artificial neural network , machine learning , engineering , robot , paleontology , geometry , mathematics , biology , mobile robot , aerospace engineering
Numerous self-driving cars algorithms rely on grid maps for motionplanning, obstacles avoidance, or environment perception. Obtained from fusedsensory information, the occupancy grids (OGs) are nowadays among the mostpopular solutions used in series production in the automotive industry. In this paper, we extend Deep Grid Net (DGN) [18], a deep learning (DL) system designedfor understanding the context in which an autonomous car is driving. We considerthis paper a granular approach to DGN method due to the improvements addedto the original research [18]. DGN incorporates a learned driving environmentrepresentation based on OGs obtained from raw real-world Lidar data and constructed on top of the Dempster-Shafer (DS) theory. Our system is able to predictin real-time if the vehicle is driving on the highway, on county roads, inside acity, in parking lots or is stuck in a traffic jam. The predicted driving context isfurther used for switching between different autonomous driving strategies implemented within EB robinos, Elektrobit’s Autonomous Driving (AD) softwareplatform. We propose a neuroevolutionary approach to search the optimal hyperparameters set of DGN. Genetic algorithms (GAs) were selected due to theirdemonstrated capabilities to evolve deep neural networks with improved accuracy and processing speed. The performance of the proposed deep network hasbeen evaluated against similar competing driving context estimation classifiers.