To Balance Competitive Games Using a Multi-objective Coevolutionary Approach
Abstract
Achieving balance in competitive games is extremely difficult: even small design asymmetries can create unfair advantages that undermine gameplay quality. We show that co-evolutionary algorithms can automatically detect and quantify game imbalance, providing multiple objective metrics that complement traditional manual design methods. Our approach employs competitive co-evolution between two populations representing opposing players' strategies, using alternating multi-objective evolution cycles, to prevent dominance and promote balanced competition. The key innovation is a novel hypervolume-based "movement" metric that measures how much each player sacrifices due to competitive pressure, directly quantifying the game balance. We illustrate our approach through three comprehensive case studies spanning unbalanced and balanced game configurations. Results demonstrate that our proposed multi-objective coevolutionary search approach quantifies imbalance in terms of unequal movement from cooperative to competitive Pareto fronts by both agents, and can successfully identifies when a balanced game has occurred. This work establishes co-evolutionary optimization as a practical tool for automated game balance assessment, with applications extending beyond games to generic competitive multi-agent system requiring fairness guarantees.
Authors: Shashank Raj, Auden Garrard, Ryan Mckendrick, Bradley Feest, Kalyanmoy Deb
Published in: Genetic and Evolutionary Computation Conference (GECCO) (2026)