Property of Mit Press: for Proofreading and Indexing Purposes Only

Robots are increasingly being used to solve real-world tasks such as vacuuming or assembly-line work in industrial applications. Controllers for these robots are typically designed by engineers following textbook guidelines. Although this methodology has proven to be very successful in such applications, it quickly meets its limitations as tasks become more complex. Collective robotic systems, where groups of robots cooperate to solve a distributed task in partially unknown environments, are an example of systems that are difficult to engineer following a classical approach (Beni 2004; Sahin 2005). This is because it is not obvious how to design controllers for individual robots that cooperate toward a common goal. Evolutionary robotics (ER) has proven to be highly successful in solving difficult or underdefined engineering problems due to its potential to automatically find simple and efficient solutions (Cliff, Husbands, and Harvey 1993; Nolfi and Floreano 2000). However, for the approach to reach its full potential in solving real-world problems, we believe that a better understanding of the influence of different factors driving evolution should be developed and summarized as guidelines. An additional step is then needed to practically use the evolved controllers in a verifiable and adaptable manner. As a starting point in constructing guidelines for the evolution of cooperative robots, we turn to the biological systems that inspired evolutionary robotics. Over billions of years, animals have evolved to solve a variety of collective tasks from navigation to collective hunting, which evolutionary biologists have studied extensively. By tapping into decades of research in biology, we explore whether the insights concerning the conditions that allow for the evolution of cooperation in nature can be translated into evolutionary algorithms that are applicable to robotic problems. For this purpose, we test the biological predictions on a preliminary robotic experiment. The results obtained from this initial study are then used as a guideline for solving a problem where we evolve a group of flying robots in simulation that must cooperate 9 Evolving Cooperation: From Biology to Engineering