Intelligent game playing with AI

The full write-up of my undergraduate thesis is available here. Here is an abstract:

Video-games contain a simulated reality not unlike ours, where the developer has control over constraints and parameters which are inaccessible in the outside world. They are generally structured around tasks that are challenging for humans, and in some cases prove to be more difficult than other activities to which our species is adapted. This makes games ideal for artificial intelligence research and benchmarking. In this project, we build an intelligent agent and test it using the General Video Game AI platform. The agent is a combination of two algorithms with an outstanding track record in decision-making and pathfinding: Monte Carlo Tree Search (MCTS) and A* search. MCTS was used by DeepMind in AlphaGo, the first intelligent agent to beat the human champion in the game of Go. A* is used widely in pathfinding problems in video-games, navigation and parsing grammars in Natural Language Processing. The hybrid agent is benchmarked against a standard implementation of MCTS.