Swarm Guidance Based on Mean Field Game Concepts

Abstract

The concept of swarm comes from the biological world. Drones gather in groups of 100 or even 1000 to fly like a flock of birds, called swarms. Swarm systems satisfy several assumptions such as decentralized controls, local information and simple platforms. Swarm systems have attractive properties such as resilience, scalability, and ease of development and implementation. Swarm techniques can perform simple tasks such as moving in a coordinated direction. The flocking behavior of a group of animals that converge by local interactions toward the same heading is an example of simple consensus for decentralized dynamic systems. However, the notion of decentralized control based on local information suffers from taking into account the overall behavior of the group. For example, in a complex environment, a swarm will adapt to the presence of obstacles and congestion reactively, whereas we would like more anticipatory control. The objective of this paper is to propose a solution based on Mean Field Game (MFG) concepts to integrate macro-level knowledge at the micro-level in decentralized flocking.