A Graph-Based Multi-Agent UAV–UGV Framework for Disaster Response and Victim Localization

The system utilises multi-sensor perception fusion, cross-agent coordination, weather-adaptive 3D path planning, victim localization with 4-class triage prediction, and uncertainty-calibrated rescue dispatch. UAVs are used for wide-area aerial perception and thermal monitoring, and UGVs are used for payload transport and manipulation, and persistent ground operation. GUARD-X was tested on the MBZIRC 2020 benchmark and six custom Gazebo disaster scenarios with 2,000 episodes per scenario, with 6-scenario cross-validation. The model yielded a 96.8% Victim Detection Rate, 47.2% higher mission completion than UAV-only baselines, a 94.7% Coverage Efficiency, and an Expected Calibration Error of 0.018. Results demonstrate the ability of GUARD-X to offer reliable, coordinated and uncertainty-aware support for autonomous disaster response.