Paper 105 Details

Title

A Predator-Prey Scenario in a Virtual Ecosystem

Full Paper Download

Download via this paper's page on the MIT Press ECAL 2015 Proceedings website.

Abstract

In this poster, we present a controller model, which controls physically the simulated creatures in a biologically inspired manner; by evolving the neural connection weights of a neural network to obtain some emerged strategies of a predator prey. Here, the resulting behaviors are obtained from a predator prey encounter in a shared environment. The resulting controllers have been evolved in accordance to the physical laws and generate motion, taking into account the Newtonian dynamics. The co-evolutionary process used in this experiment was able to successfully produce stable walking movements that allowed the predator (prey) to move towards (away from) its enemy. Our best-evolved creatures of the two species were able to emerge some defense strategies.

Video Abstract