Artificial intelligence techniques towards adaptive digital games

Tamassia, M 2017, Artificial intelligence techniques towards adaptive digital games, Doctor of Philosophy (PhD), Science, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Tamassia.pdf Thesis application/pdf 2.31MB
Title Artificial intelligence techniques towards adaptive digital games
Author(s) Tamassia, M
Year 2017
Abstract Digital games rely on suspension of disbelief and challenge to immerse players in the game experience. Artificial Intelligence (AI) has a significant role in this task, both to provide adequate challenges to the player and to generate believable behaviours. An emerging area of application for digital games is augmented reality. Theme parks are expressing interest to augment the user experience via digital games. Bringing together cyber- and physical- aspects in theme parks will provide new avenues of entertainment to customers. This thesis contributes to the field of AI in games, in particular by proposing techniques aimed at improving players' experience. The technical contributions are in the field of learning from demonstration, abstraction in learning and dynamic difficulty adjustment. In particular, we propose a novel approach to learn options for the Options framework from demonstrations; a novel approach to handle progressively refined state abstractions for the Reinforcement Learning framework; a novel approach to dynamic difficulty adjustment based on state-action values, which in our experiments we compute via Monte Carlo Tree Search. All proposed techniques are tested in video games. The final contribution is an analysis of real-world data, collected in a theme park queuing area where we deployed an augmented reality mobile game; the data we collected suggests that digital games in such environments can benefit from AI techniques, which can improve time perception in players. Time perception, in fact, is altered when players enter the state of "flow", which can only happen if suspension of disbelief is maintained and if the level of challenge is adequate. The conclusion suggests this is a promising direction for investigation in future work.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Science
Subjects Pattern Recognition and Data Mining
Adaptive Agents and Intelligent Robotics
Keyword(s) Reinforcement learning
Artificial intelligence
Dynamic difficulty adjustment
Machine learning
Monte Carlo tree search
Version Filter Type
Access Statistics: 76 Abstract Views, 13 File Downloads  -  Detailed Statistics
Created: Wed, 10 Jan 2018, 13:49:00 EST by Adam Rivett
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us