Download files
Abstract
Simulation allows human beings to project their technology into the unknown with minimum financial risk. We have determined the only way to safely investigate unknown regions of space is to use agents that operate as separate entities. Self-organising swarms are agents, with low cognitive ability, that operate together to produce significant outcomes through emergent behaviour. In investigating these systems, we use a number of strategies. One is to use evolutionary algorithms; the agent uses a gene type approach to improve its performance towards a fitness function. Agents are given a set of ‘genes’ defining their characteristics the most successful agents then breed randomly combining their genes or clonally in both cases with random mutations. This leads to a fitter population. However, in the real world, this is quite a risky process particularly if the swarm is designed to managed a utility or even a defence system. The direction that the evolution will take is controlled by the selection of the fitness function alone which almost always leads to some unpredicted results. The other problem is that the fitness one is aiming at is usually poorly defined and may well be modified during the simulation process. In general, determining the fitness involves prediction which is fraught with difficulty and usually we are left with a number of potential fitnesses to explore. Currently, we are increasing the complexity of this approach by allowing only a subset of the genes to be expressed depending on the environment. This approach can be achieved using genes regulatory control in form of epigenetics, which is the upper layer of genes, to apply the most fitness subset of the genes. An area of particular interest to us, which will be used as the example in this paper, is the investigations of Asteroids. Due to the lack of information about these space objects vehicles sent to them must be robust which drives the design towards a self-organising swarm but the nature of the agents, their required properties and success will be based on a great deal of simulation.
