Date of Award
8-2024
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Electrical and Computer Engineering (Holcomb Dept. of)
Committee Chair/Advisor
Fatemeh Afghah
Committee Member
Abolfazl Razi
Committee Member
Yongqiang Wang
Abstract
As climate-exacerbated wildfires increasingly threaten landscapes and communities, there is an urgent and pressing need for sophisticated fire management technologies. Coordinated teams of Unmanned Aerial Vehicles (UAVs) present a promising solution for detection, assessment, and even incipient-stage suppression – especially when integrated into a multi-layered approach with other recent wildfire management technologies such as geostationary/polar-orbiting satellites and CCTV detection networks. However, there remains significant challenges in developing the necessary sensing, navigation, coordination, and communication subsystems that enable intelligent UAV teams. Further, federal regulations governing UAV deployment and autonomy pose constraints on real-world aerial testing, creating a disconnect between theoretical research and practical wildfire management applications. This thesis works towards bridging the gap between theory and practice, developing a high-fidelity simulated environment to train end-to-end learnable cooperative UAV team navigation with collision avoidance. Multi-Agent Reinforcement Learning is employed to train effective team performance even under partial observability and inter-agent communication restrictions. Further, this work addresses a critical gap in existing literature to enable the learning of fully three dimensional navigation through a series of curriculum learning stages.
Recommended Citation
Hopkins, Bryce, "Training UAV Teams with Multi-Agent Reinforcement Learning Towards Fully 3D Autonomous Wildfire Response" (2024). All Theses. 4372.
https://open.clemson.edu/all_theses/4372
Included in
Controls and Control Theory Commons, Environmental Monitoring Commons, Robotics Commons, Systems and Communications Commons