DCP and VarDis: An Ad-Hoc Protocol Stack for Dynamic Swarms and Formations of Drones -- Extended Version

TL;DR

This work introduces the DCP stack and VarDis to enable global coordination in drone swarms by piggybacking coordination data onto frequent local beacons. VarDis provides a CRUD-like abstraction for globally shared variables, disseminating updates through repeated beacon payloads to achieve low latency and high reliability without routing or topology maintenance. The authors perform extensive simulations across line and grid deployments, conduct sensitivity analysis using response surface methodology, and compare VarDis against a flooding baseline and a discrete-time Markov model, showing robust performance especially in denser networks. The results suggest VarDis can be tuned to approach or exceed flooding performance while reducing overhead, with clear paths for future hardware implementation and additional optimizations.

Abstract

Recently, swarms or formations of drones have received increased interest both in the literature and in applications. To dynamically adapt to their operating environment, swarm members need to communicate wirelessly for control and coordination tasks. One fundamental communication pattern required for basic safety purposes, such as collision avoidance, is beaconing, where drones frequently transmit information about their position, speed, heading, and other operational data to a local neighbourhood, using a local broadcast service. In this paper, we propose and analyse a protocol stack which allows to use the recurring-beaconing primitive for additional purposes. In particular, we propose the VarDis (Variable Dissemination) protocol, which creates the abstraction of variables to which all members of a drone swarm have (read) access, and which can naturally be used for centralized control of a swarm, amongst other applications. We describe the involved protocols and provide a mainly simulation-based performance analysis of VarDis.

Figures (16)

• Figure 1: DCP Protocol Stack
• Figure 2: Line deployment with fixed density and $K$ nodes. The half-filled node is the sole producer of a variable, the fully filled node is the reference consumer. Each link is $L_P$ meters, where $P$ is a given packet error rate.
• Figure 3: Line deployment with variable density and $K$ nodes. The half-filled node is the sole producer of a variable, the fully filled node is the reference consumer.
• Figure 4: Grid deployment with $K$ nodes on a side and $K^2$ nodes in total, with either fixed density (all vertical or horizontal links have common packet error rate $L_P$) or variable density (total side length fixed to 1,120 m). All nodes produce a variable, the producer and consumer we focus on are highlighted as the half-filled node in the lower right corner (producer) and the fully filled node in the upper left corner (consumer).
• Figure 5: Average sequence number gap for per-hop PER $P=20\,\%$, with and without summaries, varying repCnt and beaconing rate.