Vision-Driven Autonomous UAS for Search and Rescue

Abstract

Search and rescue missions demand rapid, reliable, and intelligent action in uncertain, often degraded environments. Autonomous unmanned aerial

vehicles (UAVs) are increasingly well suited for these tasks, offering scalable coverage, high mobility, and the ability to reach otherwise inaccessible

areas. However, to be effective in real-world search and rescue operations, UAVs must navigate safely near obstacles, detect and avoid flying

vehicles and animals, localize robustly under sensor and environmental degradation, and plan informative paths in vast, uncertain spaces.

This thesis presents a narrative centered on enabling search and rescue capabilities across several robotics domains and arguing for further

development and use of vision-based autonomous UAVs in search and rescue contexts. The discussion focuses on the advantages of these systems. These include vision-based detect-and-avoid (DAA) capability,

sensor-aware path planning based on locations of interest, vision-based localization when GPS is unreliable, and human interpretability of camera
data. Also included is ongoing work on an informative path planning (IPP) system for maritime search, motivated by aerial search for lost
vessels. Collectively, these efforts highlight the practicality and utility of robotics systems in the search and rescue domain.