An Enhanced Multi-Agent System with Evolution Mechanism to Approximate Physarum Transport Networks

Abstract

The primitive unicellular organism slime mould Physarum Polycephalum has attracted much attention from researchers of both biology and computer science fields. Biological experiments have revealed that its foraging mechanism can be used to solve shortest path problems, while its foraging process can construct efficient networks among food sources. Oregonator Model and Cellular Automaton have been proposed to simulate the intelligence and morphology of Physarum. To better understand the network formation of Physarum, a multi-agent system (MAS) model of particles was introduced by Jones, which can simulate many interesting patterns of Physarum transport networks. The MAS model is improved in three aspects: the number of sensors of each individual agent is reduced to two, while the function of each sensor is extended to sample both chemical nutrient and trail. A memory module is added to the architecture of an agent, by which the evolution mechanism can be achieved to maintain the population of the system. With such improvements, the system is more flexible and adaptive, and the networks constructed using the MAS model are more approximate to the ones by Physarum in biological experiments. All these are verified by constructing stable networks including Steiner's minimum tree, cycle-like and spanning trees.