Engineering Swarming Systems

Most multi-agent systems are inspired by classical AI, whose objective was to realize humanlevel intelligence in a computer. As the field has moved toward multiple agents, there has been a presumption that individual agents still aspire to high-level intelligence. Swarming systems follow an alternative model, inspired more by artificial life than artificial intelligence. The individual agents in these systems may be non-cognitive, but complex, robust cognition emerges from their interactions. This paper defines swarming and the concepts of self-organization and emergence that underlie it. It describes the kinds of problems for which it is well suited, explores why it functions, and outlines some initial principles of an engineering methodology for developing artificial swarming systems. 1 WHAT is Swarming? We define swarming as “useful self-organization of multiple entities through local interactions.” We begin by reviewing other definitions, then focus in on organization and self-organization, and the relation of these concepts with emergence. 1.1 Swarming Definitions of swarming have been proposed by insect ethologists, roboticists, and military historians. Of the many definitions that have been proposed, a few will illustrate the main themes. Students of biological systems use it to describe decentralized selforganizing behavior in populations of (usually simple) animals [9, 10, 14, 38]. Swarming has been defined (e.g., [10]) as “distributed problemsolving devices inspired by collective behavior of social insect colonies and other animal societies.” Table 1 lists a few examples that have been studied. Table 1: Some Examples of Swarming in Nature Swarming Behavior Entities Pattern Generation Bacteria, Slime Mold Path Formation Ants Nest Sorting Ants Cooperative Transport Ants Food Source Selection Ants, Bees Thermoregulation Bees Task Allocation Wasps Hive Construction Bees, Wasps, Hornets, Termites Synchronization Fireflies Feeding Aggregation Bark Beetles Web Construction Spiders Schooling Fish Flocking Birds Prey Surrounding Wolves The use of the term to describe artificial systems can be traced to Beni, Hackwood, and Wang in the late 1980’s [4-7, 26, 27]. Their work focuses on populations of cellular robots, and they use the term to describe self-organization through 10/31/2003 12:12 PM Page 1 Engineering Swarming Systems local interactions. In the context of unpiloted air vehicles (UAV), Clough defines a swarm as a “collection of autonomous individuals relying on local sensing and reactive behaviors interacting such that a global behavior emerges from the interactions” [16]. He distinguishes swarming (resulting from reactive behaviors of simple homogeneous entities performing simple tasks) from the emergent behavior of heterogeneous teams of deliberative entities performing complex tasks. Recently, “swarming” has come into vogue in the military to describe a battlefield tactic that involves decentralized, pulsed attacks [2, 20, 21, 30]. Military historians focus less on the process of self-organization and more on the resulting organization itself: “the systematic pulsing of force and/or fire by dispersed, internetted units, so as to strike the adversary from all directions simultaneously” [2]; a “scheme of maneuver” consisting of “a convergent attack of several semi-autonomous (or autonomous) units on a target” [21]. The connection with insect applications is not coincidental. Insect self-organization is robust, adaptive, and persistent, as anyone can attest who has tried to keep ants out of the kitchen or defeat a termite infestation, and military commanders would love to be able to inflict the frustration, discomfort, and demoralization that a swarm of bees can visit on their victims. The linkage between swarming and warfare is ancient. In the Bible, God promises to demoralize the indigenous population of Canaan before the invading Israelites in the words, “I will send the hornet before you” (Exodus 23:28; cf. Deuteronomy 7:20; Joshua 24:12). In the eighteenth dynasty (1550 BC), the ancient Egyptians awarded military heroes a gold and silver medal in the form of a stylized fly ( ) [29], and there is evidence that the ancients sometimes hurled hives of stinging insects against their enemies [35]. Figure 1 Figure 1: Egyptian “fly” medal for military heroes, 1550 BC (National Gallery of Art) For the purpose of this paper, we will define swarming as “useful self-organization of multiple entities through local interactions.” This definition highlights elements of the others that have been suggested. “Useful” emphasizes that we are interested in engineering systems that are answerable to someone outside of the system boundary for their behavior. Some forms of self-organized behavior, such as riots and oscillation, might be interesting to a biologist, but undesirable in a commercial or military application. Self-organization is most prominent in the robotic definitions, since the concern there is to distinguish swarming from conventional top-down control schemes. The military definition does not emphasize selforganization, perhaps because of a historic tradition of topdown centralized control. We do not require that the selforganization result from reactive rather than deliberative individual behavior. Thus our definition includes not only Clough’s “swarms” but also his “teams,” if they meet the other terms of the definition. The notion of multiple entities is common to all definitions, and indeed is intrinsic to the common-sense use of the term. A major motivator for swarming is the proliferation of autonomous platforms, such as vehicles, communications systems, and sensor systems. Although 10/31/2003 12:12 PM Page 2 Engineering Swarming Systems these systems are often referred to as “unmanned,” in current practice it would be more accurate to describe them as “remotely manned.” The flight crew for a Predator UAV consists of two people. Housing them in a control van rather than on board the flying platform considerably reduces their risk, but does not reduce the manpower requirements for fielding the vehicle. A major promise of swarming is multiplying the number of platforms that a single person can effectively control. Our focus on local interactions has two motivations: a need and a promise. The need is a growing concern about communication congestion. The promise is the observation that local interactions suffice to maintain long-range coordination in biological systems, so that we ought to be able to reverse-engineer the underlying mechanisms for use in synthetic systems. 1.2 Organization As used in expressions such as “self-organization,” the word “organization” has at least three distinct, but related, meanings: it can refer to a mapping, a process, or a structure. Organization1 is a mapping from a system to an ordered set, e.g., Such a mapping permits us to say that one system is “more organized” than another (or than the same system at a different time). Different detailed definitions for this mapping are possible. Common themes will include entropy and symmetry, as illustrated in Figure 2. Denote a system by an upper-case letter, and its elements as the same letter in lower-case, indexed. Thus A = {a1, ..., an} and B = {b1, ..., bk} denote two systems. The entropy of a system A is denoted by S(A). With these concepts, we can meaningfully assert Org(A) > Org(B) if