Estimating Belief Updates in AI-Driven Drone Controls for Urban Search and Rescue Operations

Recent advancements in artificial intelligence (AI) have significantly enhanced the capabilities of autonomous systems in complex decision making within challenging built environments, particularly in urban search and rescue (USAR) operations, a critical component of postdisaster response. These operations require swift decision making to balance conflicting objectives, such as optimizing path planning while ensuring comprehensive environmental mapping. Current Explainable AI (XAI) approaches often focus on rationalizing or explaining the decision outcomes of AI agents but do not adequately address the methodologies needed to understand how AI systems adapt their belief systems based on real-time, limited local data, such as wind disturbances and obstacle distributions. This paper introduces an evidence accumulation model inspired by cognitive psychology, designed to estimate AI belief updates. Utilizing multiobjective reinforcement learning (MORL), this model trains drones in simulated urban settings characterized by dynamic wind fields. A reverse reasoning approach is then applied to interpret AI decisions using collected behavioral data and environmental observations. A comprehensive simulation study was conducted to assess the effectiveness of this model. The results demonstrate that the model effectively identifies how limited local information influences AI behavior after belief updates, offering a human cognitive analysis perspective to rationalize AI actions. This research contributes innovative methodologies for explaining the sophisticated cognitive flow of AI systems, laying the groundwork for improved human-AI alignment in postdisaster applications in urban settings.