Cooperative Drift Mitigation for UAV Swarms in GNSS-Denied Environments

Abstract

Unmanned Aerial Vehicles (UAVs) operating in GNSS-denied environments typically rely on Inertial Measurement Units (IMUs) for position estimation. However, this approach is susceptible to error accumulation, commonly known as inertial drift. Standard industry solutions mitigate this issue by fusing IMU data with external sensors such as LiDAR or cameras. However, these sensing modalities are not suitable for all environments. An alternative approach is to leverage cooperation within a swarm of drones, enabling agents to exchange information and improve their position estimates collectively. One such method employs a Distributed Graph Optimization (DGO) algorithm to cross-reference spatial uncertainties among UAVs in the swarm. However, existing DGO frameworks are primarily validated using relative swarm cohesion metrics, which provide little insight into the swarm's absolute positioning accuracy.

To address this limitation, this paper evaluates a basic DGO state estimation model against a basic Dead Reckoning (DR) baseline. A Python-based simulation environment was developed, and four experimental conditions were investigated: varying sensor quality, swarm size, flight duration, and trajectory geometry. The results show that DGO outperforms DR under degraded sensor conditions, whereas DR maintains lower error during short-duration flights when high-quality sensors are available. Crucially, a temporal breakeven point is identified beyond which the unbounded error growth of DR exceeds that of the cooperative DGO framework. This finding demonstrates that while standalone DR offers superior short-term precision, cooperative estimation provides a more stable and sustainable framework for prolonged operations in GNSS-denied environments.