Collaborative Localization for Fleets of Underwater Drifters Diba Mirza, Curt Schurgers diba@ucsd.edu, curts@ece.ucsd.edu University of California, San Diego 9500 Gilman Drive La Jolla, CA 92312 Abstract- A crucial goal for ocean-sensing systems is to obtain spatially-rich data that allows us to understand the correlations between ocean processes. To achieve this, we consider a system consisting of swarms of underwater drifters that float freely with currents and therefore achieve high spatial sampling of the ocean. Our goal is to ensure that this system is self-organizing and autonomous which is key for practical large-scale deployments in remote regions. We consider a crucial aspect of data collection, namely, determining the locations at which data was sensed. Given our overall design goal, we propose a localization strategy where nodes collaborate to determine their positions autonomously without using long range transponders on surface buoys or ships. This in turn significantly impacts the cost and ease of deployment of such systems. Further, we determine optimum configurations for the swarm so that their position estimation error is minimized. I. INTRODUCTION Collecting spatially rich data is the key to understanding many intricate bio-physical processes and the space-time scales at which they operate. This is because correlations between processes can be made if they are concurrently sensed over large regions at high resolutions. Traditional stand alone systems such as remotely controlled robots or AUVs are severely limited in this respect. Static deployments of sensors are also constrained in the amount of spatial data they can collect. To meet these challenges, we consider a system consisting of a swarm of free-floating underwater drifters or drogues [1]. While such drifters have been previously used as passive elements to learn the dynamics of ocean processes (e.g. by remotely imaging their trajectories), we use them to achieve a truly distributed and self-organizing data collection system for sensing phenomena in 3D [2]. To achieve this, drifters communicate over acoustic links to collaborate and coordinate their sensing, thus forming a wireless network as we have previously described [2]. Thus, various functionalities fundamental for data collection can be performed within the network, in other words the swarm can self-organize. One such essential functionality is to determine where and when data was collected which is crucial for correctly interpreting it and extracting spatial correlations. Therefore, devices in the swarm have to be localized, which will be our focus in this paper. Traditional methods for obtaining location information, such as GPS, cannot be used underwater due to high attenuation of radio waves. However, in principle, nodes can estimate their positions from distance measurements to devices with known positions. Often transponders mounted on moored buoys or ships are used to track the position of underwater devices such as AUVs. However, for the networks that we consider where drifters follow unpredictable trajectories in often long-term missions, nodes would easily go out of the transmission range of fixed transponders. Deploying sufficient number of surface elements at appropriate locations is both costly and requires much planning. In addition nodes cannot be localized by surface elements when they travel to large depths since they would no longer be within communication range. Therefore, our goal is to develop a cost effective localization strategy that does not require prior planning, infrastructure and doesn’t restrict the sensing region of the network. To achieve this, nodes in the network should collaborate to estimate their positions autonomously or self- localize. II. NETWORK SELF-LOCALIZATION: CHALLENGES We consider the application where nodes sense the spatial variation of a process in 3D. Starting at the surface where the network is deployed, drogues use buoyancy control to descend deeper into the ocean. Once a maximum desired depth is reached, they travel back to the surface and the process is repeated to obtain the spatial map of the process as shown in Figure1. z y Figure 1: 3D sensing with partial mobility control Direction of controlled motion