|dc.contributor.author||Nguyen, Joseph Luan||-|
|dc.description.abstract||The ability of UAVs to cover large areas efficiently is valuable for information gathering missions. For long-term information gathering, a UAV may extend its endurance by accessing energy sources present in the atmosphere. Thermals are a favourable source of wind energy and thermal soaring is adopted in this thesis to enable long-term information gathering. This thesis proposes energy-constrained path planning algorithms for a gliding UAV to maximise information gain given a mission time that greatly exceeds the UAV's endurance.
This thesis is motivated by the problem of probabilistic target-search performed by an energy-constrained UAV, which is tasked to simultaneously search for a lost ground target and explore for thermals to regain energy. This problem is termed informative soaring (IFS) and combines informative path planning (IPP) with energy constraints. IFS is shown to be NP-hard by showing that it has a similar problem structure to the weight-constrained shortest path problem with replenishments. While an optimal solution may not exist in polynomial time, this thesis proposes path planning algorithms based on informed tree search to find high quality plans with low computational cost.
This thesis addresses complex probabilistic belief maps and three primary contributions are presented:
• First, IFS is formulated as a graph search problem by observing that any feasible long-term plan must alternate between 1) information gathering between thermals and 2) replenishing energy within thermals. This is a first step to reducing the large search state space.
• The second contribution is observing that a complex belief map can be viewed as a collection of information clusters and using a divide and conquer approach, cluster tree search (CTS), to efficiently find high-quality plans in the large search state space. In CTS, near-greedy tree search is used to find locally optimal plans and two global planning versions are proposed to combine local plans into a full plan. Monte Carlo simulation studies show that CTS produces similar plans to variations of exhaustive search, but runs five to 20 times faster. The more computationally efficient version, CTSDP, uses dynamic programming (DP) to optimally combine local plans. CTSDP is executed in real time on board a UAV to demonstrate computational feasibility.
• The third contribution is an extension of CTS to unknown drifting thermals. A thermal exploration map is created to detect new thermals that will eventually intercept clusters, and therefore be valuable to the mission. Time windows are computed for known thermals and an optimal cluster visit schedule is formed. A tree search algorithm called CTSDrift combines CTS and thermal exploration. Using 2400 Monte Carlo simulations, CTSDrift is evaluated against a Full Knowledge method that has full knowledge of the thermal field and a Greedy method. On average, CTSDrift outperforms Greedy in one-third of trials, and achieves similar performance to Full Knowledge when environmental conditions are favourable.||en_AU|
|dc.publisher||University of Sydney||en_AU|
|dc.publisher||Faculty of Engineering and Information Technologies||en_AU|
|dc.publisher||School of Aerospace, Mechanical & Mechatronic Engineering||en_AU|
|dc.subject||unmanned aerial vehicle (UAV)||en_AU|
|dc.title||Long-term Informative Path Planning with Autonomous Soaring||en_AU|
|dc.type.pubtype||Doctor of Philosophy Ph.D.||en_AU|
|Appears in Collections:||Sydney Digital Theses (Open Access)|