Emergent Behaviour Evolution in Collective Autonomous Mobile Robots CĂTĂLIN-DANIEL CĂLEANU, VIRGIL TIPONUŢ, IVAN BOGDANOV, IOAN LIE Applied Electronics Department University “POLITEHNICA” Timişoara 2 V. Pârvan Blvd. 300223 Timişoara ROMANIA Abstract: - This paper deals with genetic algorithm based methods for finding optimal structure for a neural network (weights and biases) and for a fuzzy controller (rule set) to control a group of mobile autonomous robots. We have implemented a predator and prey pursuing environment as a test bed for our evolving agents. Using theirs sensorial information and an evolutionary based behaviour decision controller the robots are acting in order to minimize the distance between them and the targets locations. The proposed approach is capable of dealing with changing environments and its effectiveness and efficiency is demonstrated by simulation studies. The goal of the robots, namely catching the targets, could be fulfilled only trough an emergent social behaviour observed in our experimental results. Key-Words: Autonomous mobile robots, emergent behaviour evolution, neural networks, genetic algorithms 1 Introduction Collective autonomous mobile robots systems represent nowadays the subject of much research [1]-[3]. Systems of distributed robots or software agents have obvious advantages such as faster operation, higher efficiency, and better reliability than a single robot system. Novel complex tasks could be solved in parallel either through explicit cooperation, competition or some combination thereof [4]. Lots of algorithms were proposed for efficient control of a group of robots/agents accomplishing a wide variety of tasks. Among them distributed auction algorithm [5], the graph matching algorithm [6], network simplex algorithm [7]. More recent, the computational intelligence paradigms (artificial neural networks (ANN) [8], fuzzy systems (FS) [9], genetic algorithms (GA) [10], etc.) try to incarnate the unique behaviours of living creatures in nature onto artefacts like robots. Our paper exploit the combination of several soft computing techniques, ANNs, fuzzy systems and evolutionary algorithms, in the development of a pursuing system in order to demonstrate emergent characteristic of the artificial life into machine learning. The resulting evolutionary artificial neural network (EANN) and evolutionary fuzzy logic controller (EFLC) represent important tools in the evolutionary robotics domain [11]. Numerous researches about the collective autonomous mobile robot evolutionary control in the predator-prey pursuit system have been carried out. Parker and Parashkevov [12] employed a cyclic genetic algorithm (CGA) for evolving single loop control programs for legged robots. The design proved successful for the evolution of a controller that allowed a robot to efficiently search for a static target in a square area. Also they demonstrate the capability of CGAs with conditional branching to generate a controller the predator in a predator-prey scenario. Nolfi and Foreano [13] investigated the role of co-evolution in the context of evolutionary robotics simulating a pursuit system with two robots (a predator and a prey) in real environments. McKennoch et. al. [14] studies software agents in a predator-prey environment when the movements of prey agents are evolved upon a Mamdani type fuzzy inference system. It has been shown that probabilistic predation and starvation forces, along with simulated communication activity act upon agents, causing them to cluster. Analysing the literature, some of the weak points of the control paradigms could be pointed out, e.g. when the gene structure was represented by fuzzy membership functions a long time is required in order to resolve the pursuing problem, when reinforcement learning is employed there are difficulties of learning when the rewards of taken actions are not instantly computed [15]. Also an environment which is dynamically changing represents in most cases an important issue. Therefore, in this paper, to resolve those problems, we apply the EANN and EFLC paradigms for modelling the robots evolving in a dynamic virtual environment. This paper is distributed as follows: section 2 describes the architecture of the evolutionary systems used. The paper continues with section 3, describing the experimental results using our robot simulator. Finally, in section 4, we present the conclusions and future work. 12th WSEAS International Conference on SYSTEMS, Heraklion, Greece, July 22-24, 2008 ISBN: 978-960-6766-83-1 428 ISSN: 1790-2769