Paper story
Guarding a place from a swarm
Adversarial patrolling usually assumes one clever intruder. We asked a different question: what if the attackers are a swarm, and the guard is a shepherd? This is the story of a fourth-year project that became an IEEE conference paper, without the mathematics.
The problem
Many intruders, one guard
Adversarial patrolling is a classic security question: how should a guard move so that an intruder never finds an opening? The classic version is a game of wits against a single clever adversary, usually played on a grid or a graph. But some modern threats do not look like one clever intruder. They look like a swarm: many simple agents, all drawn toward the thing you are protecting, none of them individually smart, all of them persistent.
Against a swarm, the classic playbook has little to say. The attackers do not plan, so out-thinking them is beside the point. The real question becomes physical: can one mobile defender exert enough steady influence to keep a whole crowd out?
The scene: an attacking swarm is drawn toward a protected area of interest, and a single defender interposes to hold it back.
The idea
The guard is a sheepdog in reverse
This is shepherding turned inside out. A sheepdog uses repulsion to move a flock toward a goal; our defender uses the same force language to keep a flock away from one. The attackers behave exactly like sheep: they flock together, avoid collisions, and flee from the defender when it comes close. The defender exploits those reflexes with four behaviours, and a one-step look-ahead chooses whichever behaviour best keeps the swarm out at each moment.
Drive pushes the cluster away from the area; collect regroups a stray before it slips around the flank; intercept places the defender on the swarm’s approach path; patrol holds an orbit when no threat is pressing.
What we found
Reflexes are enough to hold the line
In simulation, the look-ahead defender holds the attacking swarm at a safe distance from the protected area, switching fluidly between pushing the crowd back, chasing down strays, and cutting off approaches. The deeper point mirrors our earlier work on the limits of reactive shepherding: simple force-based control goes remarkably far, provided you understand the geometry of when and where to apply it.
The story behind the paper
A fourth-year project, done properly
This paper grew out of Jonathan Zhou’s fourth-year engineering project at UNSW Canberra, which I supervised as principal supervisor with my colleague Aya Hussein. Jonathan carried the idea from a project brief to a peer-reviewed IEEE conference paper presented in Kuching, Malaysia. It is a good example of what undergraduate research can be: a real question, a defensible method, and a result the community can build on.
Cite & explore
The formal version
J. Zhou, H. El-Fiqi and A. Hussein, “Adversarial Patrolling Using a Shepherding Approach,” 2024 IEEE International Conference on Systems, Man, and Cybernetics (SMC), Kuching, Malaysia, pp. 839–844, 2024. doi:10.1109/SMC54092.2024.10832074
Where reactive shepherding breaks: the companion story →
Back to the paper gallery →
How this page was written. The research, the results, and the ideas here are mine and my co-authors’. To retell them in plain language, I worked with an AI writing assistant that helped draft the text and render the diagrams in this site’s style. I reviewed and edited everything, and the technical responsibility rests with me. If the prose reads a little differently from my papers, that is why.