Distributed control in Multi-Agent Systems: A preliminary model of autonomous MAV swarms

This article focuses on the use of multi-agent systems for modelling of micro-unmanned aerial vehicles (MAVs) in a distributed control task. The task regards a search and destroy scenario in the context of security and urban counter-terrorism. In the simulations developed, a swarm composed of four autonomous flying robots, driven by an embodied neural network controller, has to approach a target deployed somewhere within the given environment. When close enough to the target, one of the aircraft needs to carry out a detonation in order to neutralize it. The controllers used by the MAVs evolve through a genetic algorithm. The preliminary results presented here demonstrate how the adaptive evolutionary approach can be successfully employed to develop controllers of this kind. The MAV swarms evolved in this way are in fact able to reach and hit the target, navigating through an obstacle-full environment. Further works on this model will focus on the development of a 3D physical simulator, in order to move towards the usage of MAVs with neural network controllers in real applicative urban scenarios.