Evolving Intelligent Mario Controller by Reinforcement Learning

Artificial Intelligence for computer games is an interesting topic which attracts intensive attention recently. In this context, Mario AI Competition modifies a Super Mario Bros game to be a benchmark software for people who program AI controller to direct Mario and make him overcome the different levels. This competition was handled in the IEEE Games Innovation Conference and the IEEE Symposium on Computational Intelligence and Games since 2009. In this paper, we study the application of Reinforcement Learning to construct a Mario AI controller that learns from the complex game environment. We train the controller to grow stronger for dealing with several difficulties and types of levels. In controller developing phase, we design the states and actions cautiously to reduce the search space, and make Reinforcement Learning suitable for the requirement of online learning.