The Modular Agent Based Evolver framework (often referred to as MABE), is a software platform that aims to allow users to easily conduct and compare evolution experiments .
MABE is open-source and available on GitHub here, which includes a wiki with helpful information.
The core idea of MABE centers around the concept of “modules” that can be easily be swapped for one another. Modules are split into five categories: worlds, optimizers, archivists, brains, and genomes. A detailed description of what each type of module does is provided in MABE’s wiki. Modules are designed so that a user can easily substitute one module for another of the same type. For example, the user can swap a Markov Brain module for a neural network brain to see how the difference in substrate changes the experiment’s results.
MABE2, a successor to MABE, is currently being developed by Charles Ofria, Cliff Bohm, and other members of Michigan State University’s Devolab. While still early in development, MABE2 aims to solve several issues from MABE.
The current version of MABE2 can be found on GitHub here.