Autonomous UAV Swarms: Distributed Microservices, Heterogeneous Swarms, and Zoom Maneuvers
Loading...
Date
2024-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Over the last few years, precision agriculture has greatly benefited from advancements in Unmanned Aerial Vehicles (UAVs). UAVs used in crop map ping allow farmers and researchers to tailor farming practices to the specific needs of individual management zones. Yet despite their benefits, traditional exhaustive approaches to UAV remote sensing are constrained by the low battery capacities of UAVs. To optimize battery usage, we present our work across 3 papers on alternative reinforcement learning (RL) and multi agent reinforcement learning (MARL) approaches to UAV remote sensing, namely: SoftwarePilot 2.0, heterogeneous swarms, and the Zoom maneuver. Starting with SoftwarePilot 2.0., we introduce a software package that sup ports scalable autonomous UAV swarms through microservice model design, container deployment technologies, and specialized MARL policies. These improvements to SoftwarePilot 2.0. reduced energy costs by 50% and improved swarm decision-making by 2.1 times from base SoftwarePilot. Moreover, we further explored multi-agent strategies through the extrapolation of multiple health metrics with heterogeneous UAV swarms. Heterogeneous swarms can capture data from multiple types of sensors, e.g. RGB, thermal, multi-spectral, and hyper-spectral cameras, and then extrapolate for various distinct health metrics across the whole field. Our preliminary results showed 90% accuracy from extrapolation from sampling only 40% of the field. Lastly, we proposed a study on the battery and accuracy tradeoffs of Zoom maneuvers. Moreover, Zoom maneuvers or changes in altitude trade battery for increased local accuracy. Our study considers the computational battery cost and flight battery cost tradeoffs of autonomous vs. RL implementations of Zoom maneuvers. Ultimately, this paper provides new insights into the execution and performance of various autonomous and multi-agent UAV remote search strategies.
Description
Keywords
unmanned aerial vehicles, edge computing, swarm, agriculture