Automatic Step Evolution



A common issue in Artificial Life research, and mainly in open-ended evolution simulations, is that of defining the bootstrap conditions of the simulations. One usual technique employed is the random initialization of individuals at the start of each simulation. However, by using this initialization method, we force the evolutionary process to always start from scratch, and thus require more time to accomplish the objective. Artificial Life simulations, being typically, very time consuming, suffer particularly when applying this method. In a previous paper we described a technique we call step evolution, analogous to incremental evolution techniques, that can be used to shorten the time needed to evolve complex behaviors in open-ended evolutionary simulations. In this paper we further extend this technique by automating the process of stepping the simulation. We provide results from experiments done on an open-ended evolution of foraging scenario, where agents evolve, adapting to a world with a day and night cycle. The results show that we can indeed automate this process and achieve a performance at least as good as on the best performant non-automated version.


14th edition of the Ibero-American Conference on Artificial Intelligence, IBERAMIA 2014, November 2014

Cited by

No citations found