Development of Physical Agents for Robot Swarms J.M.Fernandez 1,2 , B.P.Amavasai 2 , W.F.W.Othman 2 , S.P.McKibbin 2 , F.Caparrelli 2 , J.R.Travis 2 & A.Goude 2 1 Universidad Politécnica de Valencia, Camino de Vera, s/n. 46022 Valencia, Spain. jofervai@etsii.upv.es 2 Microsystems and Machine Vision Laboratory Materials and Engineering Research Institute, Sheffield Hallam University, Pond Street, Sheffield S1 1WB, United Kingdom {b.p.amavasai, wan.a.othman, s.p.mckibbin, f.caparrelli, j.r.travis,a.goude}@shu.ac.uk Abstract – The field of swarm engineering is a growing area of research. The aim of swarm engineering is to design multi-robot platforms that are able to mimick biological robot swarms in performing tasks where a cluster of robots can excel better than just one. In designing large robot swarms, cost plays an important role. In this paper we attempt to develop a low cost autonomous platform that is modular in design. This allows the system to be expanded or extended as and when necessary. By using a limited amount of sensors and without direct communication between each platform, we are able to simulate and demonstrate a number of behaviours. Keywords: Swarm engineering, multi-robot, modular robot, behaviour-based systems, self-organisation 1 Introduction A swarm of robots is essentially a multi-agent system, albeit on a much larger scale. Robots in a swarm communicate and cooperate in order to perform or solve tasks that are beyond the capability of individual robots. The creation of small robot swarms has become a topic of great interest in recent years. As far back as 1986, Brooks[1] advocated the creation of simple robots with simple behaviours, instead of complex abstract models and interpreters to achieve intelligent behaviours. These behaviour-based models have become widely accepted within the research community due to the simplicity of the approach coupled with the flexibility and potential for generalisation to learning. In this paper, it will be shown that, by applying very simple individual rules on simply designed robotic platforms, it is possible to create swarm behaviours of increasing complexity. Ultimately, by studying the relationship between the applied rules and the attained behaviours, it should be possible, by reverse engineering, to define what rules need to be applied in order to achieve the desired behaviours in a swarm. Although being a relatively young research area, robot swarms are already being used in a variety of applications ranging from mine detection[2] to solving routing in mobile networks problems[3] to the more futuristic areas of autonomous task recognition and replication. This paper takes a practical approach to building low- complexity, low-cost, swarm-capable robot agents to complement the work carried out by Othman et. al. [4]. 2 Background A number of multi-robot initiatives currently exist. These robots fall into multiple categories, which include cost, size, autonomy and capability. The CentiBOTS [5] project at SRI aims to build 100 robots that map, track and guard unknown terrain in a coherent manner. These heterogeneous set of robots essentially consists of mini-ITX boards with either Pentium or VIA based processors. The onboard computes are connected to a variety of sensors, including cameras, laser range finders and an inertial navigation system. Due to the on-board computational power available, complex algorithms can be designed and run on these systems. Each robot can also communicate with other robots through a wireless 802.11b link. In the Swarm-bots [6] project the aim is to construct and study novel approaches to the design and implementation of self-organising and self-assembling robots, using recent theories in the area of swarm intelligence. The robots uses a differential drive systems through two tracks, each powered by three wheels. Each robot is equipped with an Intel 400 Mhz XScale processor. Sensors include an omnidirectional camera, inclinometers, and temperature and humidity sensors. The robot also uses PIC based microcontrollers that communicate and interfaces between sensors and the main processor. Each robot is equipped with two servo based grippers which can be used to hold onto other robots to form solid chains. This is useful in order to