Emergent Structure in Multi-agent Systems Using Geometric Embeddings

TL;DR

This paper presents a decentralized control method using geometric embeddings to enable multi-agent systems, like drone swarms, to self-organize into stable, energy-efficient trajectories based solely on local observations.

Contribution

It introduces a novel injective virtual embedding approach that preserves structure and stability, allowing agents to achieve coordinated trajectories without global communication.

Findings

Swarm of quadcopters successfully self-organize into desired trajectories.

The method maintains even separation among agents.

The approach is versatile and applicable to real UAV swarms.

Abstract

This work investigates the self-organization of multi-agent systems into closed trajectories, a common requirement in unmanned aerial vehicle (UAV) surveillance tasks. In such scenarios, smooth, unbiased control signals save energy and mitigate mechanical strain. We propose a decentralized control system architecture that produces a globally stable emergent structure from local observations only; there is no requirement for agents to share a global plan or follow prescribed trajectories. Central to our approach is the formulation of an injective virtual embedding induced by rotations from the actual agent positions. This embedding serves as a structure-preserving map around which all agent stabilize their relative positions and permits the use of well-established linear control techniques. We construct the embedding such that it is topologically equivalent to the desired trajectory (i.e., a homeomorphism), thereby preserving the stability characteristics. We demonstrate the versatility of this approach through implementation on a swarm of Quanser QDrone quadcopters. Results demonstrate the quadcopters self-organize into the desired trajectory while maintaining even separation.