Adaptive optimization of solution time in a distributed multiagent system

The complexity and generally unconstrained interactions among agents in large-scale, multi-agent systems make it difficult to control the time these systems require to produce a solution. An important principle for optimizing the solution time for these large-scale systems is to isolate the complexity introduced at each point in the multi-agent society. In this paper, we discuss solution time control techniques used in a large-scale society of agents that performs logistics planning and replanning in chaotic environments. We describe four techniques for controlling the quantity and timing of the information flow between agents, and show the impact of these techniques on the solution time. These four techniques are: producing a low-resolution solution before the detailed plan is complete, reducing rework by controlling the transmission of information based on local consistency, reducing information redundancy by transmitting differences, and predicting information provided by agents in the absence of communications. Finally, we discuss how these techniques can enhance robustness and survivability.