Multi-agent pathfinding for unmanned aerial vehicles
CovenanteeKokuritsu Jōhōgaku Kenkyūjo
Document typeMaster thesis
Rights accessOpen Access
Unmanned aerial vehicles (UAVs), commonly known as drones, have become more and more prevalent in recent years. In particular, governmental organizations and companies around the world are starting to research how UAVs can be used to perform tasks such as package deliver, disaster investigation and surveillance of key assets such as pipelines, railroads and bridges. NASA is currently in the early stages of developing an air traffic control system specifically designed to manage UAV operations in low-altitude airspace. Companies such as Amazon and Rakuten are testing large-scale drone deliver services in the USA and Japan. To perform these tasks, safe and conflict-free routes for concurrently operating UAVs must be found. This can be done using multi-agent pathfinding (mapf) algorithms, although the correct choice of algorithms is not clear. This is because many state of the art mapf algorithms have only been tested in 2D space in maps with many obstacles, while UAVs operate in 3D space in open maps with few obstacles. In addition, when an unexpected event occurs in the airspace and UAVs are forced to deviate from their original routes while inflight, new conflict-free routes must be found. Planning for these unexpected events is commonly known as contingency planning. With manned aircraft, contingency plans can be created in advance or on a case-by-case basis while inflight. The scale at which UAVs operate, combined with the fact that unexpected events may occur anywhere at any time make both advanced planning and planning on a case-by-case basis impossible. Thus, a new approach is needed. Online multi-agent pathfinding (online mapf) looks to be a promising solution. Online mapf utilizes traditional mapf algorithms to perform path planning in real-time. That is, new routes for UAVs are found while inflight. The primary contribution of this thesis is to present one possible approach to UAV contingency planning using online multi-agent pathfinding algorithms, which can be used as a baseline for future research and development. It also provides an in-depth overview and analysis of offline mapf algorithms with the goal of determining which ones are likely to perform best when applied to UAVs. Finally, to further this same goal, a few different mapf algorithms are experimentally tested and analyzed.
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder