Dynamic Difficulty Adjustment Using Behavior Trees

Student thesis: Master thesis (including HD thesis)

  • Torkil Olsen
  • Huy Long Phan
  • Kenneth Sejrsgaard-Jacobsen
2. term, Computer Science, Master (Master Programme)
We investigate the feasibility of adjusting the difficulty in a 2D fighting game using behavior trees. We present an overview of dynamic difficulty adjustment properties and define a fitness function as an appropriate measurement of difficulty based on a specific scenario. The purpose is to ensure a balanced game between a player and an agent by minimizing the difference between the player's skill and the difficulty of the agent. We propose two different approaches to adjust the difficulty of an agent, using behavior trees, based on the agent's fitness. The first method introduces a mechanism to switch between multiple predefined behavior trees to create variations in the difficulty. The second method utilizes a dynamic behavior tree where probabilities inherent in the tree structure are adjusted to change the behavior. Tests conducted show promising results for the second method.
Publication date9 Jun 2011
Number of pages58
ID: 52870701