Quantitative Comparison Of Two Approaches To Agent Cooperation

Multi-agent systems (MASs) are characterized by collections of autonomous agents that interact with each other in simple ways, but the collection of agents as a whole is characterized by emergent behavior (EB) which will have properties that individual agents do not.

Engineers want to design artificial MASs for a variety of reasons including military operations. The idea is for the EB to be carrying out the mission itself. In this scenario, the agents are relatively inexpensive and expendable. A large body of work exists in MAS research for military applications and a variety of different designs have been proposed. Because there is no uniform framework for communicating quantitative research results dealing with agent cooperation, it is difficult for engineers to make well-informed design decisions. We demonstrate an example of how to qualitatively and more importantly, quantitatively compare two approaches to agent cooperation. The example includes a formal experiment design and data analysis. Our example MAS deals with cooperative suppression of enemy air defense (C-SEAD). We consider two approaches to agent cooperation: state-based and Artificial Physics (AP). In this case, our example indicates that the state-based approach is better. In the corpus of research work, ideas for mobile robot MAS cooperative control, etc. are usually “validated” by simple, ad hoc simulations. We are not convinced these simulations actually validate the ideas proposed because they lack an appreciation of system-level constraints. E.g., you cannot engineer an unmanned aerial vehicle (UAV) that has three-hour endurance yet weighs less than one pound—there is no viable source of power that can do that. If one is engineering a MAS for the real-world, we think the validation of each candidate design requires high-fidelity simulation of all important aspects of the MAS—movement, communication, resource consumption, etc. This thesis demonstrates the need for such simulations.