With technology rapidly advancing, the virtual worlds are increasing in size and complexity, and they are more difficult to create. To aid the development process, procedural generation can be implemented (Houdini can be used for procedural generation, as well as other tools and plugins for game engines such as Unity and Unreal), but there are downsides to consider when doing so. These downsides arise from the loss of control of the designer: the outcome can become repetitive, out of the theme of the game, or simply unusable.
Our goal is to improve on the creation of game levels with better efficiency and increased creativity. For that purpose, we test a mixed-initiative approach on level design in games. We believe that this approach could capitalize on the benefits of procedural generation, without taking away control from the designer. Similar ideas have been tested studied before ().
How we tested it
To gather results, we implemented a tool while consulting experts in the domain to ensure its validity. The experiments consisted of individual sessions with level design students where each participant built two villages: one without the AI part of the tool, and one with AI suggestions. The experiment ended with an interview.
The tool was built as a generic system based on rules which were adjusted so that the AI could give suggestions for building a village. The rules are based on the buildings and the terrain that are planned to be used. To be concrete, there are relations between buildings (so that a barn is suggested close to farmhouses) and relations between terrain and buildings (so that a watermill is suggested close to the water). By mixing rules with negative weight for one building with itself, the “overpopulation” with a specific type of building can also be avoided. While building the world, the AI analyses the environment automatically so that it can always offer appropriate suggestions based on what is currently placed.
Results from observations and interviews show that participants found it easier to create villages using this tool and were more interested in the process because of it. The participants also stated that the tool stimulated their creativity, and if more procedural generation was integrated, it could considerably lower the time required to build levels.
The topic of procedural generation will bring further opportunities, as this research also showed: the participants stated that the concept of the tool would be useful in production. Improvements to this tool that would have a big influence on the design process are the implementation of road network generation, or AI that learns the style of the designer.
This kind of tool can have applications in other domains based on settlements: the film industry, city planning, historical simulations, and others, but it can also be used with other elements. The generic rules can accommodate placements of foliage, non-playable characters, or they can be modified to test this approach in other domains where logic AI can be implemented, besides placing elements in a virtual world.
Depending on the implementation, systems where humans are supported by AI can be used either in development, or in education. The AI can be set up with information by experts in a domain, and then be used to teach juniors with suggestions and notes on why those choices were suggested.
To conclude, we advise to further develop mixed-initiative procedural level design tools as it can effectively stimulate creativity and productivity in the game development process. Applications of this system can be introduced in other domains as well.
If you want to learn more about the research behind this article, you can access the full thesis at http://dx.doi.org/10.13140/RG.2.2.23595.92966/1.
 Emilien, Arnaud & Vimont, Ulysse & Cani, Marie-Paule & Poulin, Pierre & Benes, Bedrich. (2015). WorldBrush: Interactive Example-based Synthesis of Procedural Virtual Worlds.
 Delarosa, O., Dong, H., Ruan, M., Khalifa, A., Togelius, J. (2021). Mixed-Initiative Level Design with RL Brush. In: Romero, J., Martins, T., Rodríguez-Fernández, N. (eds) Artificial Intelligence in Music, Sound, Art and Design. EvoMUSART 2021. Lecture Notes in Computer Science(), vol 12693. Springer, Cham. https://doi.org/10.1007/978-3-...