Anticipatory control of vehicle swarms with virtual supervision

Abstract

This paper presents an application of anticipatory network theory to model the behavior of a swarm of autonomous vehicles that share a common goal. In addition, each vehicle optimizes its individual performance criterion that is subordinated to the group goal. The internal swarm organization resembles a hierarchical control system where the top level is distinguished only by the hierarchy of goals, instead of a fixed assignment of powers or permissions. The arising variable hierarchy depends on the type of momentary performance of the swarm units: those performing activities leading directly to reaching the superordinated goal have the right-of-way and priority access to shared resources. Two principal problems need to be solved in this context. The first one is to recognize temporal hierarchies by swarm vehicles. This is accomplished by ensuring appropriate communication between vehicles via a local network. The second problem is to define behavior strategies that yield the best attainment of the common goal while individual indicators are nondominated. Solving both problems ensures a balance between cooperative (reaching a shared goal) and self-interested (individual goals) behavior. Finding a compromise strategy is equivalent to solving a certain anticipatory network. This model can be applied to supervising mining vehicle cooperation, where efficient communication and coordination of individual actions play central roles.