Online Level Generation in Super Mario Bros via Learning Constructive Primitives

In procedural content generation (PCG), how to assure the quality of procedural games and how to provide effective control for designers are two major challenges. To tackle these issues, this paper exploits the synergy between rule-based and learning-based methods to produce quality yet controllable game segments in Super Mario Bros (SMB), hereinafter named constructive primitives (CPs). Easy-to-design rules are employed for removal of apparently unappealing game segments, and subsequent data-driven quality evaluation function is learned based on designer's annotations to deal with more complicated quality issues. The learned CPs provide not only quality game segments but also an effective control manner at a local level for designers. As a result, a complete quality game level can be generated online by integrating relevant constructive primitives via controllable parameters. Extensive simulation results demonstrate that the proposed approach efficiently generates controllable yet quality game levels in terms of different quality measures.