Shepherding large swarms with visibility constraints

Abstract

The shepherding problem is a problem setting in the multi-agent systems domain. The primary aim is to guide or ‘shepherd’ a swarm of unwilling agents to a destination, with the help of autonomous agents. We explore this problem domain, implementing the state of the art algorithms for both single and multiple driver settings. We present an algorithm for multi-shepherd herding which can successfully herd swarms with thousands of agents using only three drivers - outperforming previous work by more than an order of magnitude. We introduce visibility constraints and obstacles to emulate real-world conditions and achieve an extremely robust model.