Overview
Cooperative boundary tracking with teams of autonomous surface vehicles.
This thesis studies how Multi-Agent Reinforcement Learning can support in situ monitoring of marine hazards such as harmful algal blooms and oil spills. The core objective is accurate boundary tracking using decentralized agents with local sensing and limited communication.
The work introduces a flexible simulation framework with static and dynamic contamination settings, and evaluates TransfQMix-based policies across observation modes, communication constraints, sensor noise, and team-size transfer.
Highlights
Key outcomes
Emergent coordination
Agents learn cooperative perimeter coverage, zig-zag traversal, and collision-aware behavior in complex plume geometries.
Robustness findings
Policies show robustness to communication loss and noise, while transfer between static and dynamic environments remains challenging.
Deployment insight
The thesis includes an initial analysis of computational feasibility for edge inference on constrained onboard systems.
Project Resources
Interactive thesis content
These are the existing thesis pages from the original site, now grouped under this project.