Training Game Agents for Different Difficulties

Assigned Human-Computer Interaction Computer Science Games Engineering Bachelor Master Project Thesis Games Research Lab

Background

Machine learning (ML) techniques can yield artificial intelligence (AI) with super-human game-playing abilities for diverse adversarial games. Yet, for an enjoyable gaming experience an AI needs to pose a challenge but still be beatable. In fact, balancing the difficulty for novices and throughout the game is a hard challenge. All the more so, if an AI emerged by means of ML as opposed to its manual design, e.g. based on rules, behaviour trees or state machines.

Tasks

• Research methods to train agents/NPCs by means of ML
• Find or create a suitable game environment to train agents/NPCs
• Research difficulty measures and propose evaluation functions or target states for different levels of difficulty
• Train agents based on the identified formulations of difficulty levels
• Evaluate the resulting agents’ difficulties by means of a user study

Contact Persons at the University Würzburg