Learning-based game generators in Super Mario Bros

This research project aims at a novel procedural content generation method to generate high quality games in Super Mario Bros online. As a result, we have proposed a hybrid approach by exploiting the synergy between rule-based and learning-based methods to produce constructive primitives, quality yet controllable game segments in SMB. As a result, easy-to-design rules are employed for removal of apparently unappealing game segments, and then active learning by encoding a game designer's knowledge implicitly is applied to obtain constructive primitives. Thus, online level generation and real-time content adaptation can be done by integrating those constructive primitives with a variety of criteria. In addition, the constructive primitives generated by our approach should also be able to work for off-line procedural level generation for richer content by using more complex controllable parameters.

For demonstration, we have applied our approach to two automatic content generation scenarios: one is an online game generator that can generate high quality games via controllable parameters, and the other is an adaptable generator which adapts games to match challenges that a game player can cope with. The demo prototype on two scenarios including source code and executable jar files are available below. For instance, some levels generated by our approaches (the first two levels produced by the online generator and the last two levels produced by the adaptive geneator) are as follows.

If you have any questions, please contact me (shipa@cs.manchester.ac.uk).