University Of Tasmania
Thompson_whole_thesis.pdf (12.46 MB)

A mission planner to improve the cost-effectiveness of autonomous marine vehicle deployments

Download (12.46 MB)
posted on 2023-05-28, 08:56 authored by Thompson, FF
Operators and end users of Autonomous Marine Vehicles (AMVs) value mission reliability (i.e. the probability that the AMV satisfies mission goals). Operators traditionally use conservative safety margins when planning missions for AMVs to reduce the risk of uncertainty affecting mission success. Although pragmatic, the policy is largely based on user experience and runs the risk of underusing or overextending the vehicle in missions where experience is lacking. Quantifying components of uncertainty improves mission plans by adding safety margins more closely attuned to both task completion and vehicle survivability. Mission uncertainty is influenced by how well the mission plan predicts the vehicle's performance in the marine environment. Current planners formulate missions as a schedule of timed tasks and only briefly consider the performance prediction in terms of time. A potential solution is to plan with energy consumption and energy capacity as the planning constraints. Unlike the current standard marine vehicle temporal planners, energy planning anticipates loadings on the AMV as it fulfils assigned tasks. In multi-vehicle missions, plans are made by allocating and scheduling tasks to vehicles based on their energy capacity. The task schedules can be optimised to minimise energy consumption and maximise the number of tasks completed. Based on the plan's energy prediction, the measured energy consumption of deployed vehicles can be evaluated to determine if correction is necessary. Energy planning maintains survivability by adhering to vehicle battery constraints and increases the operator's confidence in the plan as it has been thoroughly quantified, meaning that the safety margins governing the vehicle's overall exposure to risk are appropriately tuned. This thesis describes the development of a two-stage AMV Energy Planning Framework (EPF). In the first stage, the EPF automates the process of formulating the planning problem from mission data specified by the operator and then solves it to provide energy efficient plans for each vehicle in the fleet. The EPF formulates the mission data into the Team Orienteering Problem (TOP) and uses Monte Carlo simulation on the vehicle dynamic model to obtain the expected energy consumption for each possible edge in the TOP graph. The solver, a discrete implementation of particle swarm optimisation, was evaluated using TOP testing data and was found to generate near optimum plans with resource scarcity (i.e. AMV fleets containing up to 4 members). The full first stage of the EPF was tested on a dataset from a wind turbine array inspection mission and was shown to generate concurrent mission plans for multiple underwater vehicles. The second stage occurs during deployment, where a mission supervisor agent onboard the vehicle monitors the progress of the plan relative to the vehicle's measured energy consumption. The supervisor decides if the current plan is likely to be completed and, if not, decides on a recourse action according to policy parameters specified by the operator. A prototype AMV was developed to perform missions provided by the first stage of the EPF and the results were used to evaluate the second stage. During deployments of the AMV in a lake environment, it was demonstrated that manipulation of the supervisor policy parameters resulted in encouragement of conservative or risky behaviours. A conservative policy caused the vehicle to return home sooner, completing tasks along the way. In contrast, a risky reliability parameter caused the vehicle to persist with the original mission plan for longer. By providing an accurate model of the vehicle and environment, operators using the EPF will be able to plan efficient missions tailored to the capabilities of their vehicles. Accurate models are not always available, so a data-driven Long Short-Term Memory (LSTM) neural network model was proposed to forecast vehicle energy consumptions based on a learned vehicle model. Conservative or risk-taking behaviour policies can be specified to influence task completion by the AMV without stranding it in an irrecoverable position. The result is a nuanced method for planning efficient and safe missions for AMVs. Operators using an EPF approach will be able to do more with their AMVs per mission, increasing the cost-effectiveness of expeditions.


Publication status

  • Unpublished

Rights statement

Copyright 2019 the author Chapter 2 appears to be the equivalent of this is the peer reviewed version of the following article: Thompson, F, Guihen, D. 2019. Review of mission planning for autonomous marine vehicle fleets, Journal of field robotics, 36:(2), 333‚Äö- 354, which has been published in final form at This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Use of Self-Archived Versions Chapter 3 appears to be the equivalent of a post-print version of an article published as: Thompson, F, Galeazzi, R., 2020. Robust mission planning for autonomous marine vehicle fleets, Robotics and autonomous systems, 124, 1-23 Chapter 4 appears to be the equivalent of this is the pre-peer reviewed version of the following article: Thompson, F, Galeazzi, R., Guihen, D., 2020. Field trials of an energy‚ÄövÑv™aware mission planner implemented on an autonomous surface vehicle, Journal of field robotics, Early view, 11 February 2020, 1-23, which has been published in final form at [ This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Use of Self-Archived Versions

Repository Status

  • Open

Usage metrics

    Thesis collection


    No categories selected