Applying Tree Search and Reinforcement Learning to Competitive and Human-Like MOBA AI
Competitive strategy games are the focus of a great deal of research in artificial intelligence and game-playing. They offer direct feedback on how well a player has performed - you either win or you lose. and there is an obvious goal to strive for - superiority over top-level human players. Moreover creating a proficient agent for complex, real-time strategy games could provide insight into how to solve general real-world problems. This has led to a decent amount of interest into research into these topics. Recently Google DeepMind and Blizzard have released an API for Starcraft 2, and OpenAI have created an agent for Dota 2 which consistently beats top players at a simplified version of the game.
Multiplayer Online Battle Arena, or MOBA, games are typically 5v5 team games in which each player has a set of complex abilities, and each team must gain resources by killing the other team and the other team’s army, with the final goal becoming strong enough to destroy the opposing team’s base. This provides a lot of disparate problems to solve simultaneously and in real-time, such as micromanagement of your player-character, planning long and medium-term strategy as the game goes on, predicting the opposing team’s strategy and much more.
I aim to explore the creation of a competitive AI for MOBA games, which has direct application both in player enjoyment in games where a player disconnects, but also perhaps for professional competitive players to be able train without revealing their strategies by playing against real players online. It may also provide useful insight to more general real-world problem solving, since the genre is so complex.
I did my undergraduate masters degree at Oxford University, in computer science. During this degree I did two projects relating to the implementation of AI for games, creating agents to play both Go and Starcraft: Brood War. Since graduating in 2015 I have spent the best part of the last 2 years as a software engineer in Bristol at a software consultancy, and a product-based start-up. I’m looking forward to returning to academia and trying my hand at game development.
Home institution: York
Ready to apply?
Once you have identified your potential supervisor, we would encourage you to contact them to discuss your research proposal.Learn More