Strategies of Division of Labour for Improving Task Efficiency in Multi-Robot Systems Sifat Momen & Amanda J.C. Sharkey Neurocomputing and Robotics Group, Department of Computer Science, University of Sheffield Regent Court, Portobello, Sheffield S1 4DP, UK {s.momen, amanda}@dcs.shef.ac.uk Abstract— This paper addresses the issue of allocating tasks in multi-robot systems and evaluates the strategies that various ant colonies display in carrying out tasks and proposes four different task allocation strategies within the realms of multi-robot systems. The paper also introduces the concept of cost-benefit ratio as a measure of performance index for the task allocation strategies identified. Experimental results show that uniformly distributed specialised workers are better than generalised workers in carrying out tasks. The paper then improves the performance of the strategies by incorporating a long term memory to carry out tasks. The proposed model for task allocation is highly efficient, accurate and consistent with the biological counterparts. The model offers benefits in designing efficient multi-robot systems that can carry out more than one task. Keywords— cost benefit ratio, division of labour, multi-robot system, swarm robotics, task efficiency I. INTRODUCTION The field of swarm robotics (SR) (for brief histories of the subject see [1] and [2]) has recently gained a swelling interest amongst researchers in areas as different as biology and engineering [2, 3] who are interested in investigating the underlying mechanisms of multi-agent systems. The popularity of this area has been manifold including the benefit it offers to multi-robot systems in terms of flexibility, scalability and robustness and also due to the platform it provides to the biologists to explore the mechanisms underlying the biological behaviours in animals. The concept of SR has close ties with that of swarm intelligence since both were initially inspired by the synchronised behaviour of eusocial insects [4] that emerges due to the local interactions between the neighbouring agents and also due to the interactions between agents and the environment in vicinity [5]. Previous works in the field of SR have been involved in carrying out simple tasks with relatively limited number of robots (see e.g. stick pulling experiments [6], cooperative transport of food items [7] and foraging of food items from the environment [8, 9]). Recently, however, there has been a lot of emphasis on developing a swarm robotic system comprising of hundreds of robots executing a number of tasks concurrently. This paper investigates the issue of task allocation in a multi-agent system so as to improve the task efficiency. Our work is strongly inspired from the behaviour of eusocial insects such as ants which provide us with a number of keen techniques for dividing the tasks among the members of the colony. The originality of the paper lies in: (1) identification of the techniques of DOL in ants, (2) developing a series of agent based models to investigate each of the techniques, (3) empirically investigate the effect of each of the techniques in (2) on the task efficiency and finally (4) in improving the algorithms so as to improve the task efficiency in the multi- agent system. The rest of the paper is organised as follows: Section 2 investigates the mechanisms of DOL in social insects such as ants. Section 3 describes the proposed models followed by the experimental results in Section 4. Finally in Section 5, we conclude the paper with an empirical comparison of the models as well as with a remark on our future work. II. DIVISION OF LABOUR IN EUSOCIAL INSECTS Ants are classified as eusocial insects belonging to the family of Formicidae of the order Hymenoptera. They are perhaps the most successful living beings that have made their marks on the earth. They are extremely small in size and weigh very little (on average between 1 and 5 milligrams) yet they live at large and weigh (total weight) as much as all human beings on the earth [10]. Ants are extremely diverse in terms of their colony size, organization of tasks and also cooperation among the nest mates for the benefit of the colony. There are currently over 12,000 known species of ants, each maintaining highly organized colonies and nests with colony size ranging from a few individuals to 20,000,000 individuals [11]. How these tiny insects are so successful in maintaining colonies of different sizes? What techniques have they embraced that enabled them to be socially so successful? Research has shown that DOL has played a significant role to their organisational success [12-14]. 672 978-1-4244-5612-3/09/$26.00 c 2009 IEEE