Robot Information Gathering for Wildland Fire Monitoring

Abstract

The monitoring of complex dynamic systems, such as those encountered in disaster response, search and rescue, wildlife conservation, and environmental monitoring, presents the fundamental challenge of how to efficiently track them with limited resources and partial observability. This thesis presents algorithms and techniques for robotic information gathering in dynamic and unstructured environments. The core research focuses on two key areas: modeling and updating the belief of a dynamic system, and efficiently creating long-term plans based on a dynamic belief to maximize information. We document our approach to creating a full robotic system that incorporates these algorithms for real-life wildland fire safety monitoring. Specifically, we detail the development of a hexacopter unmanned aerial system to track the safety of wildland fire crew. To enable effective crew localization, we developed a perception pipeline trained on our real-world thermal imagery dataset. We aim towards a closed-loop system enables the robot to perceive, update its belief, create a plan, and execute that plan to gather new observations about the world. Future work includes validation of our approach in simulation real-world experiments. Finally, we discuss future directions for our robot to perform robustly in more challenging environments.