Comparison of genetic algorithms used to evolve specialisation in groups of robots Tomassino Ferrauto 12 , Gianluca Baldassarre 2 , Gabriele Di Stefano 1 , Domenico Parisi 2 1 Dipartimento di Ingegneria Elettrica e dell'Informazione, Università dell'Aquila tomassino.ferrauto@istc.cnr.it, gabriele@ing.univaq.it 2 Laboratory of Autonomous Robotics and Artificial Life, Istituto di Scienze e Tecnologie della Cognizione, Consiglio Nazionale delle Ricerche (LARAL-ISTC-CNR) gianluca.baldassarre@istc.cnr.it, domenico.parisi@istc.cnr.it Abstract This paper investigates the role of genetic algorithms in determining which kind of specialisation emerges in decentralised simulated teams of robots controlled by evolved neural networks. As shown in previous works, different tasks may be better solved by robots specialized in a particular manner. However it was not clarified how much the genetic algorithm used might drive the evolution of one kind of specialisation or another: this is the goal of this paper. The study is conducted by evolving teams of robots that have to solve two different tasks that are better accomplished by using different types of specialisation (innate versus situated). Results suggest that the type of genetic algorithm employed plays a major role in determining how robots specialize and in most of the cases the algorithms used tend to always yield the same specialization. Only one of the algorithms tested led to the emergence of the most suitable kind of specialisation for each one of the two tasks. 1 Introduction The field of collective robotics, or multi-robot systems, is receiving an increasing attention within autonomous robotics (for extensive reviews and taxonomies of multi- robot systems and of the tasks that can be tackled through them, see [1], [2], [13]. The goal of this paper is to start to systematically study the different types of specialisations and the role of genetic algorithms in determining which kind of specialisation emerges in different environmental conditions. The robots of the multi-robot systems studied here have the following properties: (a) they have to collaborate to accomplish a common task (“cooperative tasks” are the most studied in the field, see [13]); (b) have a distributed control system (there are no “leader robots” or centralised controllers within the system, cf. [6]); (c) are guided by feed-forward memory-less neural-network controllers evolved with genetic algorithms (no learning process during the tests); (d) have no explicit communication (coordination has to rely upon perception, physical interactions, implicit communication, cf. [12]). Multi-robot systems are important for engineering purposes since they have a number of strengths if compared to single robots: (a) some tasks can be carried out only, Page 1