Roadmap for development of skills in Artificial Intelligence by means of a Reinforcement Learning model using a DeepRacer autonomous vehicle

• Published in: 2022 IEEE Global Engineering Education Conference (EDUCON) pp. 1355 - 1364
• Main Authors: Cota, Jamir Leal, Rodriguez, Jose A. Tavares, Alonso, Brandon Garcia, Hurtado, Carlos Vazquez
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.03.2022
• Subjects: Artificial Intelligence; Computational modeling; Deep Learning; Educational Innovation; Hardware; Machine learning algorithms; Professional Education; Reinforcement learning; Software; Training; Vehicle driving
• ISSN: 2165-9567
• DOI: 10.1109/EDUCON52537.2022.9766659

Using Deepracer, through experimentation and simulation, theoretical concepts can be applied to a practical application of reinforcement learning (RL) in a real-life problem. It was considered as a highly useful tool to develop many direct and transversal competences that students need to work within the field of artificial intelligence (AI). Nowadays the combination of Hardware, Computer Vision methods and Machine Learning (ML) algorithms for the development of controllers for vehicle driving automation have facilitated the development of solutions for this problem. The intention of this work is to show a Roadmap that was formulated to learn AI, ML and RL competencies required to prepare undergraduate students for the industry of this area, following a structured mostly practical learning plan using Deepracer AWS platform and local alternatives for training; and a physical vehicle as primary tools that have made an incredibly compact setup process and reduced complexity in the educational researching field related to learning autonomous vehicles (AV) software development process. The AWS DeepRacer framework includes all needed hardware based on a two front-camera vehicle for stereo vision and a LiDAR sensor, besides it provides a powerful computation computer for high performance in scale autonomous vehicles. The roadmap was executed testing and comparing different RL models over the modalities that Deepracer provides (by comparing both local and AWS console training) documentation of the execution of the generated Roadmap is shown to ensure that should be considered as a stable learning system that could be followed by college programs. Finally, models were tested on a physical track built, and limitations, considerations and improvements for this Roadmap are explained as a contribution for future work.