Electrotechnical and Computer Engineering
Vol. 35 No. 03 (2020): Proceedings of the Faculty of Technical Sciences
USING REINFORCEMENT LEARNING TO TRAIN AN AGENT FOR AUTONOMOUS DRIVING IN THE AIRSIM SIMULATOR
Abstract
With the advancement of deep learning and hardware, along with emergence of new technological challenges like software for robots and self-driving cars – reinforcement learning has become a fertile ground for researchers interested in this field. In this paper a novel approach has been proposed to solving the problem of autonomous driving in simulator, which combines techniques of computer vision and reinforcement learning to create a software agent that can drive a car successfully in a simulation. Evaluation of the agent has been done by comparing results and expected/given goal.
References
[1] Loaiacono i Cardamone, „Simulated car racing championship: Competition software manual,“ 2013.
[2] F. J, F. N, Vielwerth i T. J, „ Monte-Carlo Tree Search for Simulated Car Racing,“ 2015.
[3] Koutnik, Cuccu i Schmidhuber, „Evolving large-scale neural networks for vision-based reinforcement learning“.
[4] Loiacono, Prete, L. P. i C. L, „Learning to overtake in torcs using simple reinforcement leraning“.
[5] M. V. A. N, „Carma: A deep reinforcement learning approach to autonomous driving“.
[6] M. Spryn, S. Aditya i P. Dhawal. [Na mreži]. Available: https://github.com/microsoft/AutonomousDrivingCookbook/tree/master/DistributedRL.
[7] Microsoft. [Na mreži]. Available: https://microsoft.github.io/AirSim/docs/apis/.
[8] K. M i K. S, „Autonomous vehicle control via deep reinforcement learning,“ Master's thesis, 2017.
[9] R. Szelinski, Computer Vision: Algorithms and Applications, Springer, 2011.