By the end of this workshop, participants are expected to have a basic knowledge of how the Multi-dimensional Archive of Phenotypic Elites (MAP-elites) algorithm works and to be able to use it as a game design assisting tool.

In previous work we demonstrated the use of MAP-elites, a divergent search algorithm, as a game design assisting tool. The MAP Elites algorithm allows illuminating the games pace instead of just finding a single game setting via objective based optimization. In this way, different variants of the same video game (with different parameters) can be played, resulting in diverse ways of playing the game (different behaviours).

A Rolling Horizon Evolutionary Algorithm agent (RHEA) was used to automatically test each point in the game space, obtaining its performance and behaviour with that particular set of parameters. This allowed mapping the game space to the desired set of features in the behaviour space. By using MAP-Elites to search the game space, designers could define the behaviour characterization, which results in a diverse set of games, where different play styles are required to achieve the game’s objective.

In our previous work, the proposed method was applied to the 2D game cave swing. The relationship between the game space and two main parameters were explored: the agent’s average height and the average game duration. The solutions found by the MAP-elites algorithm shows that the agent’s performance distribution depends on both of the behaviour features selected.

Who should attend?

Participants should have some basic knowledge of Python and Java programming and should bring a laptop with these installed.

Workshop Organisers: Adrián Barahona-Ríos, Adam Katona, Nuria Peña Pérez and Martin Balla