Practically optimal UAV mission planning under uncertainty
File(s)
Author(s)
Qian, Qiuchen
Type
Thesis
Abstract
While Unmanned Aerial Vehicle (UAV) technologies have transformed industrial applications, the practical utility of current mission planners is limited by oversimplified assumptions about operational costs and spatial constraints. The central challenge remains in developing planners that can effectively handle energy consumption uncertainty and spatial optimization requirements while maintaining computational efficiency for online deployment on resource-constrained platforms. This thesis addresses this gap, focusing on UAV-enabled wireless power transfer and communication, by showing that combining probabilistic energy estimation with spatial-aware approaches enables robust mission planning under real-world operational constraints.
We develop two complementary frameworks. First, to manage energy uncertainty, we formulate an Uncertain and Dynamic Orienteering Problem (UDOP), introducing the Rapid Online Metaheuristic-based Planner (ROMP) that combines first principle analysis and wind field segmentation, and then advancing it with an ADaptive Approach for Probabilistic paThs (ADAPT). This framework employs a Bayesian method for real-time energy consumption estimation, adapting to dynamic factors like wind. Our experimental results demonstrate that ADAPT achieves a 100% mission success rate across all tested scenarios while maintaining comparable solution quality and computation time.
Second, for spatial optimization, we propose the Close Enough Orienteering Problem with non-uniform neighborhoods (CEOP-N) and solve it with the CRaSZe-AntS algorithm. This hybrid metaheuristic features a Randomized Steiner Zone Discretization (RSZD) scheme that identifies overlapped sub-regions for Particle Swarm Optimization (PSO) to refine continuous waypoint positioning, while an Ant Colony System (ACS) optimizes the discrete visiting sequence. Results show CRaSZe-AntS significantly outperforms single-neighborhood strategies, increasing prize collection by an average of 140.44% and reducing computation time by 55.18%. We extend this approach with CRaSZe-AntS-3D to efficiently handle non-convex geometric constraints in air-to-ground communication scenarios, achieving near-optimal solutions with only a 0.38% deviation from the optimal energy cost.
We develop two complementary frameworks. First, to manage energy uncertainty, we formulate an Uncertain and Dynamic Orienteering Problem (UDOP), introducing the Rapid Online Metaheuristic-based Planner (ROMP) that combines first principle analysis and wind field segmentation, and then advancing it with an ADaptive Approach for Probabilistic paThs (ADAPT). This framework employs a Bayesian method for real-time energy consumption estimation, adapting to dynamic factors like wind. Our experimental results demonstrate that ADAPT achieves a 100% mission success rate across all tested scenarios while maintaining comparable solution quality and computation time.
Second, for spatial optimization, we propose the Close Enough Orienteering Problem with non-uniform neighborhoods (CEOP-N) and solve it with the CRaSZe-AntS algorithm. This hybrid metaheuristic features a Randomized Steiner Zone Discretization (RSZD) scheme that identifies overlapped sub-regions for Particle Swarm Optimization (PSO) to refine continuous waypoint positioning, while an Ant Colony System (ACS) optimizes the discrete visiting sequence. Results show CRaSZe-AntS significantly outperforms single-neighborhood strategies, increasing prize collection by an average of 140.44% and reducing computation time by 55.18%. We extend this approach with CRaSZe-AntS-3D to efficiently handle non-convex geometric constraints in air-to-ground communication scenarios, achieving near-optimal solutions with only a 0.38% deviation from the optimal energy cost.
Version
Open Access
Date Issued
2025-02-01
Date Awarded
2025-09-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Boyle, David
Publisher Department
Dyson School of Design Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)