One of the great attractions of multiagent systems is their decentralization. Instead of a single centralized decision function that becomes overloaded as the problem size grows, each agent makes autonomous decisions based on its local circumstances. As the problem grows, one simply adds more agents, and the locality of their interactions permits the addition of computing hardware to allow virtually unlimited scalability.

This seductive benefit has a drawback: the more localized agent decisions become, the easier it is for the agents to become uncoordinated with one another, yielding solutions that are suboptimal or even globally inconsistent.

One of the most promising approaches to achieving multiagent coordination while preserving local decision-making is the distribution of information over the environment in the form of a field. Inspired on the one hand by the pheromone fields used by many social insects, and on the other by the physical fields in models of gravitation and electromagnetism, these computational fields are both modified and sensed locally by individual agents. The environment that supports the fields provides three functions, whose precise form depends on the application. It may aggregate them across multiple agents (a form of information fusion), propagate them (communication), and attenuate them over time (a constant-time approach to truth maintenance).

Field-based methods were introduced as a computational mechanism more than two decades ago, and have become extremely popular in a wide range of applications. For over ten years, the Agents and Pervasive Computing Group at the University of Modena and Reggio Emilia has been advancing this paradigm, and now two leaders of that group have produced a systematic exposition of the field-based approach to coordinating multiple agents. While naturally featuring examples from their own research, the authors discuss a wide range of other research.

After a brief introductory chapter, chapters 2 and 3 motivate the technique by describing the ever-increasing scope of computational systems, and the growing difficulty of coordinating across such systems using traditional methods of direct control, shared data, or event-driven programming. Chapter 3 introduces a case study that serves as a motivating example throughout the later chapters: an information environment for a museum that can support functions such as the ad hoc rendezvous of visitors, load leveling across exhibits, route planning customized to the interests of individual visitors, and emergency services.

Chapter 4 is a broad survey of field-based computing, including both a wide range of engineered systems and an analysis of several examples from the biological world. Chapter 5 develops a formal, unified model for applying these techniques. The central elements of the model are the various computational fields (co-fields) that agents maintain and sense in the environment, and the coordination field, a virtual field computed by each agent from the co-fields that it senses in its neighborhood and that guide its behavior. This virtual field allows agents to generate more complex behavior than would be possible simply by following gradients on the co-fields themselves.

Chapters 5 and 6 discuss implementation environments that support field-based coordination, including both off-the-shelf approaches and Tuples On The Air (TOTA), an infrastructure developed by the authors’ laboratory.

Chapters 8 to 10 demonstrate the potential of the field-based approach by presenting high-level designs, and in some cases citing specific systems, for three classes of problems: content-based information access, self-assembling mobile robotics, and “the cloak of invisibility,” a system of micro-sensors and display units that continuously self-organize so that someone looking at the front of an object sees a replica of what is behind it, canceling occlusion and effectively rendering the object invisible.

This volume is an excellent summary of the state of research to date; a concise guide for new adopters of field-based coordination; and a stimulating survey of potential research areas. It will be valuable both as a textbook for advanced undergraduate courses in multiagent systems and as a reference for active researchers in the field.