Collaborative Autonomy of Multi-Agent Aerial Systems for Future Disaster Response

Effective disaster response is increasingly challenged by the growing frequency and severity of natural and human-induced disasters worldwide. In such scenarios, response efficiency is primarily constrained by unreliable and severely delayed communications, inaccessible and hazardous terrain, limited resources and real-time situational awareness, which collectively hinder timely decision-making and coordinated operations. To address these challenges, this thesis investigates intelligent control, cooperative navigation, resource allocation, and agentic decision-making for multi-agent aerial systems. Chapter 1 provides the research background and motivation for developing collaborative autonomy in multi-agent aerial systems for disaster response applications. Chapter 2 reviews the state of the art in several areas, covering formation control, swarm path planning, resource allocation, and LLM-based methods. Chapter 3 develops distributed time-varying formation control protocols under dynamic communication topologies with multiple delays and computes the maximum tolerated delay for reliable coordination. Building on this foundation, Chapter 4 proposes a distributed 3D cooperative path-planning framework for large-scale UAV swarms in cluttered environments. A hybrid neighbor-forecasting strategy using a lightweight Transformer combined with constant-velocity prediction is developed, along with a learned local descriptor for directional crowding quantification and entropy-based trajectory evaluation to maintain diversity and alleviate congestion. Chapter 5 introduces a fully decentralized UAV-assisted mobile edge computing framework consisting of a K-Grouped Soft Actor-Critic with Trajectory Planning (KGSAC-TP) algorithm for UAV and a User Switching Mechanism (USM) for heterogeneous users. Furthermore, Chapter 6 designed an Agentic Retrieval-Augmented Generation framework that enables knowledge-augmented perception, situational awareness, and goal-oriented multi-robot coordination in complex disaster scenarios to generate a comprehensive report for human experts to support disaster response decision-making. Finally, Chapter 7 concludes the thesis by summarizing the key findings and discussing potential directions for future research. This thesis provides novel strategies combining control theory, artificial intelligence (AI)-driven policy, multi-agent deep reinforcement learning, and agentic AI for collaborative multi-agent aerial systems in disaster response.<p></p>

Author dc:creator

• Jia Wu (21042131)

• Multi-agent aerial systems
• AI-driven policy
• Deep reinforcement learning

Statement dc:rights

• All rights reserved
• Open Access after 2027-10-07

Identifier

10779/exe.31916157.v1

OAI identifier oai:identifier

oai:figshare.com:article/31916157