Harmonic Potential Fields: an Effective Tool for Generating a Self-Organizing Behavior

This work provide a proof-by-example of the ability of harmonic potential fields (HPF) to exhibit a self-organizing behavior that can be utilized in building decentralized, evolutionary, multi-agent systems. It is shown that the strong relation the single agent HPF approach has to the evolutionary artificial life (AL) approach may be utilized at the multiagent level to synthesize decentralized controllers that can be applied to a large variety of practical problems. We first provide a background of the single agent HPF approach along with its relation to the AL approach. Different multi-agent, HPF-based methods are presented along with simulation examples to demonstrate the utility of these techniques. Humans have long attempted to bridge the gap between actions under their direct command (control variables) and directly inaccessible desired aspects of the environment they want to influence. This is carried-out by constructing a chain of causality linking the two together; hence making those directly inaccessible aspects indirectly accessible to the human operator. The process that realizes this chain of causality is called a servo-process. There are more than on type of problems that a servo-process have to rectify in order to enable causality to flow from the control side to the desired outcome side. The failure could be caused by insufficient quantity of effort that is being exerted at the control variable side. It may be the result of incompatibility of the control effort with the aspects of the environment that is to be influenced. The lack of organization in terms of the proper spatialtemporal distribution of the assets comprising the servo-process is a serious and difficult to detect source of failure. The sufficiency of the level of information available to constructor of the servo-process is also a fundamental cause of failure. Attention in this chapter is paid to the third type of failure concerning the faulty organization of the servo-process resources. Any servo-process must, among other things, regulate the interaction among its sensory, processing, communication and actuation components. There are a number of distinct modalities in which these components are governed each suited to tackle a certain situation. Each one of these modalities gives rise to a family of planners. A planner is an intelligent, goal-oriented, context-sensitive controller that instructs the servo-process on how to deploy its actuators of motion so that a target situation may be reached in a constrained manner. Probably the most common modality used by a servo-process is the: know-plan-Act modality which is commonly called the: model-based approach (figure 1). Here, the servoprocess uses its sensors to collect data about the situation it has to deal with. This data is converted into a representation. The representation is processed to generate a plan or sub-