Browse Books

Reviewer: H. Van Dyke Parunak

Three thousand years ago, King Solomon, fascinated by the ability of insects to organize themselves without elaborate hierarchies, exhorted his readers, “Go to the ant; consider her ways, and be wise.” Recently, computer scientists have applied his advice to the design of computational processes, and along the way have clarified how insects actually accomplish their work. This book brings together a wide array of such techniques in a format that makes them clearly accessible to both computer scientists and biologists. Chapter 1 surveys self-organization in social insects, and presents general methodological guidelines, including the distinction between modeling (seeking to understand the behavior of a natural insect colony) and design (borrowing techniques from natural systems for application to engineered ones). Chapter 8 is a balanced epilogue that highlights the promise of mimicking insects in artificial systems, and also enumerates hurdles that research must overcome if these techniques are to enter the engineering mainstream. Each of the six intervening chapters focuses on a different insect behavior, describing it informally, modeling it mathematically, and suggesting how that formal model can be applied to the design of a useful artificial system. Chapter 2 describes ant foraging behavior. Ants lay down chemicals called pheromones as they move about in search of food. Other ants follow existing pheromone trails, reinforcing them in a way that yields an optimum network of paths among nests and food sources. Engineers have applied this mechanism in two ways. Ant colony optimization is a static optimization technique that maps the problem to be solved (for example, the traveling salesman problem or the quadratic assignment problem) to a graph on which a solution is a shortest path, then sends artificial ants to explore the graph. In the second application, routing in telecommunications networks, network packets lay down paths of synthetic pheromones that guide subsequent packets through the currently least congested path in the network. Chapter 3 discusses the ability of an insect colony to assign and reassign members to different tasks (such as foraging or caring for the brood), depending on current needs. The basic mechanism is a response threshold that determines whether an individual responds to a stimulus to do a certain task. These thresholds can be changed by environmental conditions, thus implementing a form of learning. The chapter illustrates possible applications of these mechanisms in two contexts. Mechanically identical robots can be mass-produced to reduce their cost, yet perform different tasks. Package delivery agents can dynamically apportion the delivery area among themselves to increase their efficiency. Chapter 4 builds on the tendency of some ant species to sort the contents of their nests, grouping dead ants into localized cemeteries or larvae into nurseries. An effective model of this behavior posits that the probability that an ant picks up an item that it encounters decreases as the number of similar items it has recently seen increases, and that the probability of dropping a carried item increases with the number of similar recently seen items. Synthetic versions of this behavior are useful in exploratory data analysis (to cluster similar items) and in graph partitioning. The chapter also describes robotic implementations of the method for clustering and sorting physical items. Chapter 5 considers how the existence of a template (a pattern in the environment, such as a temperature gradient, the body of a queen, or a brood pile) can control the location of a self-organizing activity (such as the construction of the walls of a nest). Different models are invoked to explain these behaviors, including reaction-diffusion dynamics in the case of termites responding to the pheromones emitted by a queen, and a modification of the mechanisms of chapter 4 in the case of ant nest construction. Template techniques can be used to constrain the structure that emerges in the clustering and graph partitioning processes developed in chapter 4. Chapter 6 focuses on the processes by which insects coordinate their nest-building activities. The fundamental dynamic is “stigmergy,” or coordination of activities through a shared environment, and may be either continuous (based on quantitative distinctions in environmental stimuli) or discrete (based on qualitative differences in the environment). This chapter exhibits models for discrete stigmergy, and discusses applications to self-assembling machines, a particularly important capability for micro- or nano-scale mechanisms that are difficult to manipulate with external mechanisms. Chapter 7 is inspired by the ability of teams of ants to carry items even though each ant's share of the weight is greater than the ant could carry by itself. In this case, the first formal models of the behavior come not from studies of insects, but from robotic systems for cooperative box-pushing. A subsumption architecture with layers for finding the goal, following other robots, slowing to avoid rear-end collisions, moving toward the goal, and avoiding frontal collisions yields teaming activity that moves large objects toward the goal. In spite of the limitations acknowledged in the epilogue, this book is a valuable source of examples for both biological modelers and engineering designers. The presentation, while clear and accessible, presents sufficient mathematical detail to permit readers to implement the behaviors described, and the interdisciplinary synthesis enables readers of any background to follow the discussion. The book includes an integrated bibliography of 337 items through 1998, and an index with about 750 entries.