Intelligent Pathfinding Through Collapsed Structures Using Deep Learning for Post-Disaster Operations

Rapid access to collapsed buildings is critical for post-earthquake response. This study proposes a dynamic, risk-aware pathfinding framework integrating UAV-based damage mapping, blocked-road analysis, and a Markov Decision Process (MDP) model. Collapsed buildings and debris-obstructed road segments were extracted from high-resolution UAV imagery and incorporated into the road network. Accessibility was updated using simulated field-based changes in road status, while segment risk was defined from debris extent and road width. The MDP model restored feasible access to all collapsed buildings, although some routes remained medium or high risk, demonstrating the value of dynamic geospatial decision support for post-disaster operations.