Porównanie efektywności Double DQN i neuroewolucji dla prostych problemów sterowania

Abstract

The aim of this article is to present the project about the use of Deep Q-learning and neuroevolution used by agent to learn how to control a lander for OpenAI Gym library, in LunarLander-v2 environment. The beginning of the article concerns theoretical knowledge about machine learning, neural networks, reinforcement learning and the algorithms used by agent to learn how to control the lander. Next, the article is about the effects resulting from work on the project: Lander’s environment, the script written in Python capable of learning how to control lander.