Dynamic Difficulty Adjustment Using Behavior Trees
Authors
Olsen, Torkil ; Long Phan, Huy ; Sejrsgaard-Jacobsen, Kenneth
Term
2. term
Education
Publication year
2011
Submitted on
2011-06-09
Pages
58
Abstract
Vi undersøger, om sværhedsgraden i et todimensionelt kampspil (2D) kan justeres ved hjælp af behavior trees, som er en måde at strukturere en spil-AI’s beslutninger på. Vi giver et overblik over dynamisk sværhedsgradstilpasning og definerer en fitness-funktion, dvs. et tal, der måler sværhedsgrad i et bestemt scenarie. Formålet er at skabe en balanceret kamp mellem en spiller og en computerstyret agent ved at minimere forskellen mellem spillerens niveau og agentens sværhedsgrad. Vi foreslår to måder at justere agentens sværhedsgrad på med behavior trees, baseret på agentens fitness: Den første metode skifter mellem flere foruddefinerede behavior trees for at skabe variation i sværhedsgraden. Den anden metode bruger et dynamisk behavior tree, hvor sandsynligheder indlejret i træets struktur justeres for at ændre adfærden. Test viser lovende resultater for den anden metode.
We investigate whether the difficulty in a 2D fighting game can be adjusted using behavior trees—flowchart-like structures that organize an AI opponent’s decisions. We provide an overview of dynamic difficulty adjustment and define a fitness function, i.e., a numerical measure of difficulty for a specific scenario. The goal is to keep matches balanced by minimizing the gap between the player’s skill and the computer-controlled agent’s difficulty. We propose two ways to tune the agent’s difficulty with behavior trees, based on the agent’s fitness: The first method switches between multiple predefined behavior trees to create variation in difficulty. The second method uses a dynamic behavior tree, where probabilities built into the tree are adjusted to change behavior. Tests show promising results for the second method.
[This abstract was generated with the help of AI]
Documents