Flooding can rapidly make conventional road maps unreliable, forcing emergency responders to make routing decisions across a changing and only partially observed road network. Existing research commonly treats flood forecasting, road condition sensing, evidence fusion, verification, and emergency routing as separate tasks, providing limited support for continuous decision making when information is missing, outdated, or conflicting. This concept paper proposes HydraPath, an integrated, uncertainty-aware multi-agent decision-support framework that maintains road segment accessibility beliefs by combining recursive flood forecast priors with traceable evidence from environmental sensors, unmanned aerial vehicles (UAVs), and targeted human verification. Forecasts provide accessibility priors where recent direct observations are unavailable or Outdated, while multimodal evidence is assessed according to its reliability, freshness, uncertainty, and consistency. Efficient graph coupled updating uses relationships between road segments while preventing indirect evidence alone from confirming that a road is open. The framework identifies decision critical uncertainties, allocates verification resources, and generates conservative rescue routes with auditable explanations. It also examines persistent sensor placement and adaptive sensing to improve monitoring in priority areas and reduce repeated verification requirements. The framework will be assessed through controlled FloodCastBench event replay using baseline comparisons and component ablations. Evaluation will focus on road state accuracy, uncertainty calibration, false open risk, verification efficiency, computational performance, and route safety. The expected contribution is a reproducible and transferable closed loop framework for uncertainty aware road accessibility assessment and emergency routing during floods.
Team:
Ashidu Dissanayake
Sehan Dissanayake
Shanil Diwakar
Dasun Illangasinghe
Supervisors: Prof. Dulani Meedeniya