Constrained novelty search: A study on game content generation

摘要: Novelty search is a recent algorithm geared toward exploring spaces without regard to objectives. When the presence of constraints divides space into feasible and infeasible space, interesting implications arise regarding how novelty explores such spaces. This paper elaborates on problem constrained proposes two algorithms which within both space. Inspired by FI-2pop genetic algorithm, maintain evolve separate populations, one with individuals, while each population can use its own selection method. The proposed are applied generating diverse but playable game levels, representative larger procedural content generation. Results show that two-population methods create, under certain conditions, more sets levels than current search, whether or unconstrained. However, best contingent particularities operators used. Additionally, enhancement offspring boosting shown enhance performance in all cases search.