https://repository.esi-sba.dz/jspui/handle/123456789/472
Title: | Deep Reinforcement Learning for Autonomous Driving |
Authors: | HAMMA, RAhma BOUMARAF, MAlak |
Keywords: | Autonomous Driving Reinforcement Learning Deep QNetworks Machine Learning Artificial Intelligence Deep Learning |
Issue Date: | 2023 |
Abstract: | ABSTRACT : Intelligent transportation systems (ITS) and artificial intelligence (AI) are spurring us to pave the way for the widespread adoption of autonomous vehicles (AVs), which have led to the emergence of smart roads, mobility comfort, and improved traffic safety. Many new artificial intelligence-based technologies are being used to perform important functions such as decision making, motion planning, scene understanding, vehicle control, and social behavior. A highly intelligent decision-making system that can handle complex road geometry and effectively follow routing information has become a valuable welcome asset to the world. This paper emphasizes deep reinforcement learning-based methods; which enable complex policies to be learned in high dimensional environments, as opposed to traditional planar techniques that have been widely researched in the past. The DRL algorithms used so far find solutions to the four main problems of autonomous driving, in our thesis we highlight the current challenges and point to possible future research directions. *** Resume : Les syst`emes de transport intelligents (ITS) et l’intelligence artificielle (IA) nous incitent `a ouvrir la voie `a l’adoption g´en´eralis´ee des v´ehicules autonomes (VA), qui ont conduit `a l’´emergence de routes intelligentes, au confort de la mobilit´e et `a l’am´elioration de la s´ecurit´e routi`ere. De nombreuses nouvelles technologies bas´ees sur l’intelligence artificielle sont utilis´ees pour ex´ecuter des fonctions importantes telles que la prise de d´ecision, la planification des mouvements, la compr´ehension des sc`enes, le contrˆole des v´ehicules et le comportement social. Un syst`eme de prise de d´ecision hautement intelligent capable de g´erer une g´eom´etrie routi`ere complexe et de suivre efficacement les informations d’itin´eraire est devenu un atout pr´ecieux et bienvenu dans le monde. Cet article met l’accent sur les m´ethodes bas´ees sur l’apprentissage par renforcement profond ; qui permettent d’apprendre des politiques complexes dans des environnements de haute dimension, par opposition aux techniques planaires traditionnelles qui ont fait l’objet de nombreuses recherches dans le pass´e. Les algorithmes DRL utilis´es jusqu’`a pr´esent trouvent des solutions aux quatre principaux probl`emes de la conduite autonome. Dans notre th`ese, nous mettons en ´evidence les d´efis actuels et indiquons de possibles directions de recherche futures. |
Description: | Encadrant : Pr. Sidi Mohammed BENSLIMANE / Co-encadrant : Dr. Mohamed Walid ATTAOUI |
URI: | https://repository.esi-sba.dz/jspui/handle/123456789/472 |
Appears in Collections: | Master |
File | Description | Size | Format | |
---|---|---|---|---|
Master_final-1-1.pdf | 79,33 kB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.