An AI controller for Infinite Mario Bros using evolution strategy

The Mario AI Benchmark software based on Infinite Mario Bros which is in turn, a public domain clone of Nintendo´s classic platform game Super Mario Bros. Competitions that have been held during 2009 and 2010 based on successive versions of the Mario AI Benchmark have received considerable attention and reasonable submissions. A rule based reactive controller trained using [μ+λ] evolutionary strategy has been developed within the rule set of the Mario AI competition using the Mario AI Benchmark. The controller developed is then compared to rule based controllers that have not gone through evolutionary process and thus find that evolution alongside heuristics is able to increase the chances of success in this platform games environment. The learning curve of the controller is steep but learning quickly still surpasses the basic rule-based and hardcoded controllers.