Deep Learning Algorithms for Smart Cars: A Survey

Mostafa Abotaleb^*1, Ehsaneh khodadadi², Nadjem Bailek ^3,4

¹ Department of System Programming, South Ural State University, 454080 Chelyabinsk, Russia

²Department of Chemistry and Biochemistry, University of Arkansas, Fayetteville, AR 72701, USA.

³Energies and Materials Research Laboratory, Faculty of Sciences and Technology, University of Tamanghasset, Tamanrasset, 10034, Algeria.

⁴Sustainable Development and Computer Science Laboratory, Faculty of Sciences and Technology, Ahmed Draia University of Adrar, Adrar, Algeria

Emails: abotalebmostafa@bk.ru; ekhodada@uark.edu; bailek.nadjem@univ-adrar.edu.dz

Abstract

The rate of progress in autonomous car technology has increased exponentially over the past decade, mostly thanks to advancements in deep learning and artificial intelligence. This work aims to summarize recent progress made in the application of deep learning techniques to the problem of autonomous driving. First, we will go through the deep reinforcement learning paradigm and other AI-based solutions for autonomous driving, such as convolutional and recurrent neural networks. Algorithms for driving scene recognition, path planning, behavior arbitration, and motion control were developed with these techniques in mind. Both the End2End system, which immediately converts sensory input into steering commands, and the modular perception-planning-action pipeline, each module of which is built using deep learning techniques, are the focus of our studies. We also discuss the modern challenges of building AI systems for autonomous driving, such as making sure they are safe to use, finding good places to practice, and creating effective computing hardware. This survey's comparison sheds light on the pros and cons of AI and deep learning approaches to autonomous driving, which aids in making design decisions.

Keywords: Artificial Intelligence; Deep learning; Self-driving; End2End system.