Proceedings of the International Multiconference on Computer Science and Information Technology pp. 103–120 ISSN 1896-7094 c  2007 PIPS Open-ended Evolution in Flocking Behaviour Simulation Halina Kwasnicka, Urszula Markowska-Kaczmar and Marcin Mikosik Wroclaw University of Technology, Poland halina.kwasnicka, urszula.markowska-kaczmar@pwr.wroc.pl Abstract. In this research we tried to apply open-ended evolution to breed controllers for artificial organism which would be able to man- ifest flocking behaviours. In comparison to the previous work in this area the artificial world created in our system is characterized by sig- nificant diversity of organisms. Animals were equipped with double and dynamic sight, which were suggested by other authors in their works. In experiments, many different behaviours were observed, which were very similar to those in nature, for instance: the escape of herbivore from predators, making herbivores route towards plants or a pursuit of predators after herbivores. Another interesting behaviour was grouping of predators around plants, where the probability of meeting herbivore is greater than in other places. The most advanced behaviour was creation of flocks, which was the goal of experiments. The observed motion of animals looked natural. However, to the full success it was necessary to apply steered evolution. 1 Introduction It has been ages since people have been admired flocking behaviour of animals. They have been fascinated by a veering formation of birds or a shoal motion. With the development of computer techniques they try to simulate them in computer animations but it is not an easy task for graphic artists. A smooth motion needs many subsequent frames, drawing of which is very arduous and laborious work. Additionally, each correction of route of an individual needs to take into consideration many other individuals which makes the problem more difficult. That is why this problem has been studied by many researchers. Traditional approaches are based on simulations of mathematical equations of motion ([5, 2, 9, 14]). Their construction usually requires advanced mathematical skills and exact knowledge about the studied animal’s behaviour. One of the first advanced other solution in this area was proposed by [10]. It is the main way of solving a simple flock (as well as herds and schools) simulation so far. In this approach each individual is implemented as independent element called boid. Its behaviour is controlled by rules. They determine the behaviour of a given individual on the basis of its environment (position, direction of motion of other member of flock and information about obstacles). Local interactions cause the flocking behaviour on the global level (of a whole group). In order 103