A repository for the semestral project in the course Advanced Artificial Intelligence for Games on TEK SDU Odense
The project is focused on using AI to play simple games via screen-capture technology, drawing data directly from the screen and in real time. Main focus is on the chrome dino game, which, albeit simple, still makes the AI training hard by implementing a lot of randomness, each run having a different, generated track.
The technologies used are convolutional neural networks (pytorch backend), specifically ResNet and SqueezeNet and Deep Reinforcement Learning (see the dqn branch).