Robust Belief-Reward Approximate Value Iteration for Dynamic Detection Systems Allocation in Uncertain Counter-Drone Operations*

In the realm of counter-drone operations, adversaries introduce a spectrum of drone types and quantities across various sectors of the defended area. The high maneuverability inherent to drones leads to a dynamic and unpredictable evolution of threats within these regions. Concurrently, detection systems are subject to environmental influences, which introduce variability and incompleteness in the information they gather. To address these challenges, this study presents an enhanced Partially Observable Markov Decision Processes (POMDPs) model, tailored for the complexities of drone countermeasures. The proposed model integrates a robust belief-reward frame-work and employs an approximation iteration method to refine strategic responses. This approach is designed to optimize the distribution of detection resources in countering drone threats. The experimental simulations illustrate a significant improvement in the precision of detection system allocation, achieving an enhancement of approximately 30% over standard methods. Additionally, the methodology proposed herein streamlines the dynamic deployment process of defensive systems, reducing the time expenditure in such operations.