Decentralized Declustering of Multiple Underactuated Autonomous Surface Vehicles

Abstract

Multi-agent systems have seen a significant rise in research interest, enabled by the increasing availability of low-cost autonomous platforms and motivated by a wide range of emerging applications. However, the coordinated deployment of large numbers of autonomous vehicles in marine environments remains a nontrivial and high-risk problem, yet it is often overlooked in the literature. These vehicles are typically deployed from a single location, and their underactuated nature, close proximity, and susceptibility to external disturbances make it difficult to achieve a mission-ready configuration without collisions. In this thesis, we address the problem of transitioning a set of underactuated Autonomous Surface Vehicles (ASVs) from arbitrary and inconvenient initial conditions to a deconflicted set of deployed vehicles. We propose a decentralized and scalable method that calculates and assigns target positions to the vehicles, generates optimal paths that comply with minimum turning radius constraints, and ensures collision avoidance between the vehicles through a shared speed policy. Contributions also include a formal definition and quantification of clustering and declustering in multi-agent systems. The approach is implemented using the MOOS-IvP autonomy framework, and performance is evaluated through simulation with up to \(64\) vehicles and extensive field trials with eight vehicles. Results demonstrate that our approach reduces the time to decluster for the most challenging initial conditions by 50% compared to the current manual method. By improving efficiency and robustness while eliminating human involvement, this work streamlines ASV fleet deployments, enabling more scalable multi-agent field operations.