39th Southeastern Symposium on System Theory Mercer University Macon, GA, 31207, March 4-6, 2007 Flocking for Heterogeneous Robot Swarms: A Military Convoy Scenario Christopher JR. McCook and Joel M. Esposito Abstract- In this paper we apply a popular flocking control algorithm to a heterogeneous swarm of robots. The flocking controller contains separation, cohesion and velocity matching terms and has been shown to converge properly in the case of homogeneous swarms. The swarm models a military convoy unit with supply vehicles, defenders, and attackers. Each class of vehicle possesses different maneuvering capabilities and different objectives. In each case we modify the controller accordingly and analyze the emergent behavior of the swarm. I. INTRODUCTION warm control is a rapidly growing field in robotics. The premise of swarm robotics is that there are numerous advantages of using a group of relatively simple robots to accomplish a task over using a single highly complex robot. In swarm robotics, control is decentralized by assigning to each swarm member a simple control method that, when integrated into the full swarm, will result in higher-level swarm performance that appears to be based on a highly- sophisticated control system. The performance of the swarm that results from linking the swarm robots together is known as emergent behavior. In swarm control, there are two major fields into which swarms can be categorized, heterogeneous and homogeneous. Homogeneous swarms are those made up of robots that are completely identical, both physically and with respect to their control laws. To date, there has been a great deal of work done with homogeneous swarms and their control methods (see [1], [2], [3], [4], and [5] and the references within for detailed literature review), but there has been considerably less attention devoted to the field of heterogeneous swarms. Heterogeneous swarms combine different kinds of robots that each have their own particular strengths. When these robots work together, the swarm displays the sum of its individual units' strengths (see [6] and [7]). The motivation for this control design project is military in nature, but the concepts developed could be used in a myriad of applications, both military and civilian. Currently, military convoy operation is extremely hazardous due to the Christopher McCook is currently a midshipman at the United States Naval Academy, Annapolis, MD 21402 USA (Email: m. Joel Esposito is with the Weapons and Systems Engineering Dept., United States Naval Academy, Annapolis MD, 21402 USA (Phone: 410- 293-6135, Email: i s , supported by ONR grant num. N000 1405WR2039 1). extensive use of Improvised Explosive Devices (IED's). The thrust of the control design project was to lay the ground work for an autonomous convoy that can transport supplies to needed areas without putting human lives at risk. The convoy should be able to drive itself to a given destination while avoiding obstacles, avoiding collisions with other convoy units, maintaining convoy consolidation, and responding to enemy ambushes. In addition the convoy needs to be able to operate in areas far from centralized command stations, where teleoperation becomes unfeasible due to communication constraints. Furthermore, due to the hostile nature of military operational environments, the convoy needs to be able to overcome the loss of individual members and as a result, must not rely on a single "brain" robot to coordinate the swarm. To increase the convoy's performance, the swarm will be heterogeneous employing two types of robots, a supply unit and a defender unit. The goal of this paper is to build upon past research into successful homogenous swarm techniques and apply them to a homogeneous setting and investigate the resulting emergent behaviors. II. SWARM MODEL Let x, = [x,, y, T and V, = [VX, y denote the position and velocity of robot i 1.N. The dynamics of the robots are given by the second order system: uXi where U = [U, U u is a bounded force input vector for each robot. The basic control strategy is fashioned after the work on homogeneous swarms in [1]. The control inputs applied to the members ofthe swarm will be based on two basic parts, Ui =a, +a, one from an artificial potential field specific to the type of robot encoding a variety of position-based objectives, a, = vo and one universal velocity matching term ai= 1 NE Nvi, that contributes to swarm cohesion and alignment. A visual representation of the inputs taken from [1] illustrates the result ofthe sum oftwo control inputs on a single robot in Figure 1. 1-4244-1051-7/07/$25.00 ��2007 IEEE. 44 MA1.6 26