Author: Grzegorz Ziółkowski
Keywords: artificial neural networks, reinforcement learning, Python, Keras, PyGame, Deep Q-Learning
This thesis describes the project aimed at creating and training an artificial neural network model for playing Snake in a reinforcement learning paradigm. The network models were trained on four different types of input data, representing a combination of two parameters: coordinates (absolute/relative) and vision (presence/distance). The models were able to adapt well to the environment when trained on a small board, which can also be seen in the test results. On larger boards, they did not achieve such high and satisfactory results, which leaves the way open for further improvements and development of the project.