BRAWLERAGD

Please use this commit to playtest, as it is the last stable release. The current version is being used for experimentation in sound/mechanic generation and is unstable.


META


DATE: OCTOBER 11TH, 2022

TAGS: MULTIPLAYER, AI, RESEARCH, DESKTOP, ACTION

CONTRIBUTIONS: CORE ENGINE, UI/UX, RESEARCH PLATFORM, FIGHTER/MOVE GENERATION.

ACKNOWLEDGMENTS: Ross Mawhorter contributed the level generation and collaborated on writing the evolutionary algorithm with me.


ABOUT


BrawlerAGD looks like a simple, retro platform-fighter on the surface. Beneath the hood, however, there is a sophisticated evolutionary algorithm composed of game design documents, thorough parameterizations, agents, data analytics, and automated playtesting. The goal of BrawlerAGD is to use a genetic approach to "evolving" a fighting game (new characters, new moves, new levels, all with different physics properties). It does this by using a parametrized document that defines the game engine and elements, randomizing it, then running through a long-running search to find games that a fitness function finds "balanced".

This isn't to say that there's no fun here - playtesting showed that all participants had a good time. Excited jumps, smiles, laughters, and shouts of disappointment indicated that such a PCG system is capable of creating entertaining artifacts. Beyond that, simply tuning and watching the evolutionary tuning is entertaining on its own. Watching hundreds of games, rooting for one NPC or another, seeing a polished game develop - it's like watching a designer fumble their way through to a balanced game.


TUTORIAL


Detailed installation and gameplay information is located on the Github. If you just want to play some games, I recommend trying out the study to play games used during my experimentation with the game. If you're feeling adventurous, you can try out the evolution system - just be warned, it is a very long running simulation! I let mine run for 12-36 hours to see where I got.