A simulator of underwater glider missions for path planning

Highlights

• SeaExplorer glider trajectory simulator.

• Modeling in a time-varying marine environment.

• Experimental validation.

• Vehicle routing problems.

Abstract

In this article, we present a mission simulator developed for the Alseamar SeaExplorer underwater gliders. By taking into consideration a 4D time-varying environment, it provides an estimation of the most important metrics: the battery level, the mission duration and the distance traveled by the glider.

The main strengths of this simulator are first its use upstream as a tool to aid the glider pilots to define a feasible mission plan. Secondly, through its interoperability with the Alseamar mission management system “GLIMPSE”, it works as an internal model (replanning) during the mission. Finally, it generates adjacency matrices of weighted graphs on which our high level path planning algorithms for a single glider as well as a fleet are based.

This simulator is compared to a real experimental mission in order to confirm its accuracy and efficiency.

Introduction

Autonomous Underwater Gliders (AUGs) were invented in order to address the growing need for measures performed in marine environments on large spatial and temporal scales. They were designed to be easy to use, scalable and relatively “cheap” compared to other underwater vehicles. The concepts of underwater gliders were defined by H. Stommel in 1989 (Stommel, 1989). Their original goal was to collect water column data profiles with wide spatio-temporal coverage (thousands of kilometers and weeks to months of endurance). Thanks to buoyancy changes, they glide up and down, alternating between ascent and descent cycles. Technological breakthroughs in low-power electronics, batteries and sensors with extended capacities of satellite geolocation have paved the way to the development of the first four commercial underwater gliders: “Slocum” (Webb et al., 2001), “Spray” (Sherman et al., 2001), “Seaglider” (Eriksen et al., 2001) and “SeaExplorer” (Fommervault et al., 2018). In the balance of this article, we will focus on the latter, developed by Alseamar.

Nowadays, the gliders are classified under the category of Autonomous Underwater Vehicle (AUV). A glider is profiled for hydrodynamics in order to perform long endurance missions. According to Meyer (2016), the nose cone embeds the payload whereas the rear part is dedicated to the navigation control. Finally, the tail is a foldable mast that allows the glider to communicate its position and data to/from a terrestrial basis when it is surfacing (around every three hours).

The SeaExplorer can be customized with a large variety of sensors. The aim is to respond to the different marine applications that can be targeted: observation of coastal, meso- and submesoscale dynamics, mixing processes and transport of water and energy, impact of glider data assimilation on ocean models, acoustic detection of biological and geological activity, sediment transport/resuspension (see Meyer (2016) for an overview). The large scale of these oceanographic phenomenons also explains the growing interest in the use and the coordination of fleets of underwater gliders (Lekien et al., 2008, Leonard et al., 2010, Alvarez and Mourre, 2014, Barbier et al., 2019). The main objective of using a glider fleet is to provide the best possible and up-to-date view of an operating zone and to provide data that can be assimilated by ocean models and thus improve forecasts. The glider trajectories can be predetermined but can also be spatially refined or responsive either to sensor measurements or to marine environmental changes. Thus, it is useful to test these trajectories in simulation to assess the relevance of a mission plan for a fleet of gliders during the planning phase but also during the mission itself for the supervising.

Gliders operate at relatively low speed (around a half knot) which makes them sensitive to water currents. It is the reason why it is crucial to develop a glider mission simulator that can take into consideration a 4D environmental model (3D in space in addition to 1D in time). Most recent research works on the underwater gliders focus on the development of path planning algorithms taking into consideration a time-varying environment (Lan et al., 2022, Cai et al., 2021, Ji et al., 2022, Liu et al., 2022, Lan et al., 2021) and on the formation control and the coordination of multiple underwater gliders (Wen et al., 2022, Ma et al., 2022). In our case, the longer-term objective is to make the operating system of the gliders more autonomous.

The existing state-of-the-art simulators dedicated to robotics can be classified into three main categories, depending on the type of environment that is considered: mobile ground, aerial and marine. Most of them address both mono and multi-vehicle problems. They also emulate different types of data acquisition from many sensors and can be coded in different languages (the most popular ones being C++, Python and Java).

They are introduced by complexity order. First, in mobile ground robotics, the environment is largely known and can be imported from digital elevation models, SDF meshes, OpenStreetMap. The most famous simulators are Gazebo, CoppeliaSim (Rath et al., 2018, Tai et al., 2017, Chen et al., 2017), Webots and CARLA (Codevilla et al., 2017, Zhang et al., 2018, Dosovitskiy et al., 2017). Notice that Webots and Gazebo are used both for ground (Winkler et al., 2018, Bellicoso et al., 2018-07, Takaya et al., 2016, Zhao et al., 2015, Juang and Yeh, 2018) and aerial robotics (Koenig and Howard, 2004, Schmittle et al., 2018, Imanberdiyev et al., 2016, Mahdoui et al., 2019, Michel, 2004). Two other simulators are specifically dedicated to aerial robotics: AirSim (Shah et al., 2017) and Flightmare (Song et al., 2020). They allow to run the simulation of high-fidelity environments (warehouses, forests, etc.). Finally, concerning the marine environment, it remains the most complex one because it focuses on a relatively unknown and highly fluctuating environment. Three simulators can be mentioned: UUV Simulator (Manhães et al., 2016), UWSim (Prats et al., 2012), StoneFish (Cieślak, 2019). These simulators compute the full dynamics and hydrodynamics coefficients of autonomous underwater vehicles. The SeaExplorer model has not been implemented in these simulators so far. Thrusters, sensors (such as DVL, pressure sensors, USBL, sonars, acoustic communication device, cameras and so on) and robot arm can be simulated. They also simulate complex and realistic underwater environments such as currents, waves, seabed, lakes, etc. Until now, they are not dedicated to path planning and vehicle routing and do not provide the important metrics to be considered in such problems (mission duration, remaining energy and travel distance).

The main objective of this article is to introduce a new full mission simulator for SeaExplorers gliders in a time-varying large scale environment. This is an ongoing problem as highlighted by recent publications such as Phoemsapthawee et al. (2013) and Grande et al. (2021). There is also a real need for this kind of simulator in order to help the SeaExplorer’s pilots to prepare the mission (offline mode) but also during the execution of the mission (online mode). Indeed the gliders may operate over several weeks/months through different types of water current or different water densities which affect the course of the mission.

The simulator that we developed takes into account:

• Environmental data from the Marine Copernicus database (Marine Copernicus database, 2015),

• The flight profiles of the SeaExplorer gliders,

• A mono and multi-vehicles configuration depending on the kind of undertaken mission,

• The output of high level path planning algorithms (i.e. Hamiltonian path Rahman and Kaykobad, 2005 for example, etc.)

and provides:

• Information and forecast on consumption using models,

• The interoperability with the Alseamar mission management system GLIMPSE (GLIder Mission Piloting SystEm Besson et al., 2019). This means a precious decision aid for the end-users i.e. the SeaExplorer’s pilots during both the mission preparation (path planning and choice of a feasible mission plan) and the progress of the mission (internal model and replanning),

• A metrics report through a human–machine interface,

• When weighted graph (Bondy and Murty, 1976) based path planning algorithms are considered, it is possible to use the simulator to build the adjacency matrix (measure of the cost between any points and their neighbors taking into account either the distance or the travel time or the energy consumption),

• A comparison between the different mission plans provided by high level path planning algorithms but also between the different scenarios depending on the navigation profiles and the embedded sensors.

It is very interesting to have such a tool to reduce the cost and the time dedicated to the mission preparation phase. The simulator also needs to be scalable in order to easily integrate simulated data acquisition for real time adaptive behavior.

The article is organized as follows. After an introduction, the second section is dedicated to the presentation of the Alseamar SeaExplorer gliders and their current operating process. In the third section, the simulator requirements are presented. Then, in the fourth section, the framework and the different levels of the simulator are detailed. In the fifth section, computer simulations are performed to illustrate the behavior and relevance of our simulator. An experimental validation is also carried out through the comparison with a real experimental mission. In the last section, an evaluation of the outputs of a path planning algorithm through the simulator is presented. Finally, a conclusion is drawn and perspectives are delineated on the multi-vehicles path planning generation.