Front delineation and tracking with multiple underwater vehicles

Abstract

This study describes a method for detecting and tracking ocean fronts using multiple autonomous underwater vehicles (AUVs). Multiple vehicles, equally spaced along the expected frontal boundary, complete near parallel transects orthogonal to the front.Two different techniques are used to determine the location of the front crossing from each individual vehicle transect. The first technique uses lateral gradients to detect when a change in the observed water property occurs. The second technique uses a measure of the vertical temperature structure over a single dive to detect when the vehicle is in upwelling water. Adaptive control of the vehicles ensure they remain perpendicular to the estimated front boundary as it evolves over time. This method was demonstrated in several experiment periods totaling weeks, in and around Monterey Bay, CA, in May and June of 2017. We compare the two front detection methods, a lateral gradient front detector and an upwelling front detector using the Vertical Temperature Homogeneity Index. We introduce two metrics to evaluate the adaptive control techniques presented. We show the capability of this method for repeated sampling across a dynamic ocean front using a fleet of three types of platforms: short-range Iver AUVs, Tethys-class long-range AUVs, and Seagliders. This method extends to tracking gradients of different properties using a variety of vehicles. K E Y W O R D S adaptive sampling, autonomous underwater vehicles, multiasset planning, ocean front tracking 1 | INTRODUCTION Space-based remote sensing can provide extensive information about ocean dynamics. However, remote sensing information is generally limited to measuring the ocean surface. To probe the ocean interior efficiently requires marine vehicles such as autonomous underwater vehicles (AUVs), gliders, profiling buoys, surface vehicles, and ships sampling in situ. Unfortunately, building, deploying, and operating these in situ marine robotic explorers is expensive. As a result, any actual study involves a limited number of marine vehicles, especially when compared to the vast expanse of the ocean. Determining where to deploy and operate marine assets is a challenging problem given the 4D spatiotemporal variations in oceanographic phenomena.The use of autonomous marine vehicles will increase as the size of ocean observing systems expand to study the impact of the oceans on Earth's climate and ecosystems. The day-to-day operations of these systems will become increasingly difficult if human intervention is required. To enable large observing systems to operate more efficiently, techniques for autonomous control of assets based on science goals and data sources such as in situ J Field Robotics. 2019;36:568-586. wileyonlinelibrary.com/journal/rob

Collaborative Automation and IoT Technologies for Coastal Ocean Observing Systems

“…The m2m communication is shaped to achieve machine-consensus on task allocation and sequencing between platforms based on their metadata profiles and constraints. Collaborative robotics and m2m communications have enabled adaptive sampling (Branch et al, 2019) and extended operations (Lima et al, 2019) in coastal areas, provided an abstract mission planning paradigm.…”

Synchronous-Clock Range-Angle Relative Acoustic Navigation: A Unified Approach to Multi-AUV Localization, Command, Control and Coordination

“…These deployments were performed as independent stages in the overall experiment, where assets were deployed each day and retrieved, their collected data analyzed in a central command room where the following mission was planned and the assets redeployed. Another example of experimental work in which multiple AUVs and gliders were essentially programmed to operate independently via a central command was recently demonstrated by Branch et al (2019) in May of 2017. In these experiments, each vehicle operated independently using an adaptive behavior to detect temperature fronts; upon surfacing, data uploaded to the central command was used to update an estimate of the front location, which was then used to transmit commands to the vehicles to transect this new front estimate.…”

The Pluggable Distributed Resource Allocator (PDRA): a Middleware for Distributed Computing in Mobile Robotic Networks

“…At its core, the proposed approach relies on a task allocation mechanism that maps processing tasks to agents. A number of such task allocation algorithms for networked vehicles have been proposed in the literature within the AI planning, scheduling and robotics communities [5], [6], [7], [8], [9], [10], [2]. However, existing task allocation algorithms generally require tight integration with the agents' scheduler/executive to implement the proposed task allocation; conversely, PDRA "plugs in" to existing single-agent scheduler/executives, enabling easier integration in existing systems.…”

“…For networks with time-varying communication links that do not exhibit periodicity, we advocate for the use of the MILP-based algorithm presented in [2], which captures arbitrary time-varying communication networks at the price of increased computational complexity. A variety of other task allocation approaches are also available, including auction algorithms [8], [9] (although capturing bandwidth and throughout constraints in auction algorithms is challenging) and AI planners [5], [6], [7], [10]. We developed a set of agent-based simulation and supervision tools to assess the performance of PDRA.…”