Video Game Procedural Content Generation Through Software Transplantation

Software transplantation generates new piece of software by reusing existing parts from a given piece of software (i.e., host) to enhance other parts of a same or different software (i.e., donor). In this paper, we argue that software transplantation can be used for automatically producing video game content. We propose the first search-based algorithm for procedural content transplantation and empirically evaluating it in an industrial case study in collaboration with the developers of the commercial video game Kromaia. Specifically, our proposed approach, dubbed Imhotep, enables developers to choose what video-game content to transplant and where, and automatically searches for an appropriate solution to integrate the organ into the host. Such a search is performed by using an evolutionary algorithm guided by a simulation-based fitness function, which is novel w.r.t previous transplantation work generally guided by test-suite compliance. We empirically evaluate the effectiveness of Imhotep to transplant procedural content, specifically non-playable characters, for the commercial video game Kromaia and benchmarked it against a state-of-the-art approach in search-based procedural content generation, as well as a variant of Imhotep itself guided by a test-suite-based fitness function. Using Imhotep, Kromaia developers were able to transplant 129 distinct organs taken from the game's scenarios into five different hosts, thus generating a total of 645 new transplanted non-playable characters for this game. Moreover, we found that the game content generated by using Imhotep was 1.5 times superior to the one obtained by using its test-suite-based variant, and 2.5 times superior than the one generated by the state-of-the-art benchmark. Finally, a focus group with game developers indicated their satisfaction with the content generated by Imhotep and their willingness to use it for game development.