Thanks for the memory: Cooperative autonomous agent search in uncertain environments

摘要

The effects of cooperation between autonomous electronic or physical agents are widely studied in computational science literature. We concentrate on a homogenous population of agents in a multi-agent system (MAS) to explore the effects of useful memory on goal achievement. We use simulations to consider two-dimensional planar surfaces upon which N targets are randomly scattered. N agents exist each with a maximal interest in one specific target. Agents may observe the positions of “uninteresting” targets in the environment and communicate this information to other agents encountered within the environment. The benefits of cooperation can be approximated by pure probabilistic analysis for theoretical search success, but the introduction of real-world cost factors (e.g. fuel, energy, transmission time) associated with movement within the environment renders these predictions unusable. In pure probabilistic terms, higher numbers of cooperative agents can greatly increase search effectiveness. In systems where positive costs are associated with search, internal agent memory factors can allow agent density to approximate pure probabilistic effectiveness. Practical applications for this research include real-time electronic document search, problems in robotic multi-agent systems (e.g. “foraging” or “consumption” problems), and network coverage for wireless communication devices.