Automated Game Balancing using Evolutionary Algorithms EVOLUTIONARY ALGORITHMS

Automated Game Balancing using Evolutionary Algorithms EVOLUTIONARY ALGORITHMS BC. JURAJ KONTUĽ

Defining balance • How often has a game entity impacted player’s chance of winning the game • Some tools in the game might not be used at all, some always • How interesting, while also uncertain, a game is during its play • People’s understanding of games is highly subjective

Bayer’s Definition “Game balancing is the process of systematically modifying parameters of game components and operational rules in order to determine satisfactory configurations regarding predefined goals. ”

Predefined goals Goals aim to bring the game state closer to ‘balance’ • Change in win-rate using certain tool, or tool combination • Change in viability of certain tool, or tool combination

Other forms of game balance • Game environments generated based on ‘balance description’ [Mahlmann, Volz] • Games generated using other game’s elements [Mahlmann Dominion]

Experiments • Ms Pacman as proof of concept • Star. Craft as main subject • Goal: modify specific parameters to achieve defined state of balance • Fitness distribution:

Ms Pacman Played using simple rules-based agent • Originally 19% win-rate • Goal: 50% win-rate with minimum parameter changes • Win is considered as having score >= 1500 • GA modifies movement speed of entities

Ms Pacman

Star. Craft • Star. Craft Terrain bot vs ZZZKBot [Chris Coxe] • ZZZKBot utilizes aggressive strategy against Terrans • GA modifies attack and HP values of Terran SCV & Marine, and Zergling • Goal 1: 0% win-rate for ZZZKBot with minimum parameter changes • Goal 2: 50% win-rate for ZZZKBot with minimum parameter changes

Star. Craft

Why GA? • Excellent in finding interesting, and often innovative ways of solving tasks that do not require perfect solutions • Multiple solutions for one balance problem