Skip to main navigation menu Skip to main content Skip to site footer

Electrotechnical and Computer Engineering

Vol. 35 No. 01 (2020): Proceedings of the Faculty of Technical Sciences

USING REINFORCEMENT LEARNING TO TRAIN AN AGENT FOR THE CarRacing-v0 ENVIRONMENT

  • Novica Šarenac
DOI:
https://doi.org/10.24867/06BE05Sarenac
Submitted
December 21, 2019
Published
2019-12-21

Abstract

This paper presents training and evaluation of the agent for autonomous driving in OpenAI Gym environment CarRacing-v0. Environment is a top-down view of racing track. Agent is trained using reinforcement learning techniques. Algorithms are compared in terms of results achieved in the environment, training time and implementation details. Algorithms that are implemented and evaluated are: Deep Q-Network (DQN), Advantage Actor Critic (A2C) i Asynchronous Advantage Actor Critic (A3C).

References

[1] V. Mnih, K. Kavukcouoglu, D. Silver, A. A. Rusu, J. Veness, M. G. Bellemare, A. Graves, D. Hassabis „Human-level control through deep reinforcement learning“
[2] V. Mnih, A.P. Badia, M. Mirza, A. Graves, T. Harley, T.P. Lillicrap, D. Silver, K. Kavukcouglu „Asynchronous Methods for Deep Reinforcement Learning“
[3] https://gym.openai.com/envs/CarRacing-v0/ [pristupljeno 7.9.2019.]
[4] S. Ruder „An overview of gradient descent optimization algorithms“
[5] D. P. Kingma, J. L. Ba „Adam: A Method for Stochastic Optimization“