Design Distributed Control and Learning Algorithms for a Team of UAVs for Optimal Field Coverage
AuthorPham, Huy Xuan
AdvisorLa, Hung M
Computer Science and Engineering
StatisticsView Usage Statistics
Optimal field coverage problem refers to an active research branch that studies how we can use a finite set of sensors, such as camera, to optimally cover a field with arbitrary shape that can either be static or dynamically change over time. The problem arises in a wide range of applications, notably wildfires and oil spill tracking, military surveillance, and agriculture monitoring. In these applications, it is of growing interest to send a team of Unmanned Aerial Vehicles (UAVs) acting as a mobile sensor network, as they can provide sensing information with low costs and high flexibility, compared with traditional static monitoring methods. In this thesis, we addressed the problem by two distinct approaches: one is model-based and the other is model-free. In the first part, we proposed a model-based control framework for UAV teaming to monitor and track a dynamic field like wildfire spreading. Wildfire is well-known for their destructive ability to inflict massive damage and disruption. We characterized the optimal sensing coverage problem to work with a changing wildfire environment. We proposed a decentralized control algorithm for a team of UAVs that can autonomously and actively track the fire spreading boundary in a distributed manner. The UAV team can also effectively provide full coverage of the field and avoid in-flight collisions. Moreover, based on the proposed algorithm, some of the UAVs can automatically adjust their altitude to increase the image resolution of the border of the wildfire, while the whole team tries to maintain a complete view of it. In the second part, we utilized a model-free learning algorithm to solve a problem of optimal coverage for a static field of arbitrary shape. The objective of the UAV team is not only to fully cover the field of interest, but also to minimize overlapping among field of views of the UAVs to increase image resolution and the efficiency of the team. Because the shape of the field is unknown, traditional approaches that rely on an accurate mathematical model of the field may fail. It is thus promising to address the problem with a model-free approach. We proposed a model-free Multi-Agent Reinforcement Learning (MARL) algorithm that combines Correlated Equilibrium strategy in Game Theory and Function approximation techniques to effectively overcome challenges in MARL such as the complex dynamics of the system and the curse of dimensionality. From studying the two distinct approaches, we will draw some insights in solving optimal field coverage problems regarding each approach.