An evolutionary metaheuristic algorithm to optimise solutions to NES games

Abstract

Recently, it has been shown that lexicographic orderings and time travel can be used to automate the play of Nintendo Entertainment System (NES) games. In this work, we present a method for optimizing solutions to NES games. Since many of these classic Nintendo games are NP-hard, we propose a metaheuristic algorithm that works by borrowing operators from evolutionary algorithms. By using a search based heuristic, the algorithm is able to create basic solutions to the games and then iteratively improve upon them until it converges towards a local maximum. The optimum game solutions found by this algorithm are shown to be competitive to human players and are close to the best known times achieved by them.