Master's Thesis: Learning to Play by Watching and Knowing

Submitted by Mohamed Behery on 25. March 2019 - 16:29

Recent work in the field of game AI has been leaning towards creating agents able to play games and compete with human professionals. There exists several approaches to build such agents. Whether they are built by hand or by hard coding or using some learning techniques such as Reinforcement Learning (RL) or Genetic Programming (GP). However, Such approaches are based purely on observations.

This thesis is going to create a hybrid model that integrates observations together with rules or external knowledge about the game. This creates an agent able to cope with dynamic game rules or game goals that are frequently updated. In this thesis we're going to explore RL and GP approaches and how we can adapt these approaches to changing goals. We're also going to take a look at different approaches to high level agent behavior modeling.

For more info contact: