UAV-UAV Interception Control Design Using K-Mean Clustering

This paper introduces a real-time, clustering-driven interception framework that is designed to protect unmanned aerial vehicles (UAVs) from multiple attacking UAVs. At a centralized base station, we integrate k-means clustering with a sliding-mode backstepping controller that is implemented on defender quadrotors. K-means mitigates the multi-target issue by categorizing M attackers into N spatial clusters and offering dynamically updated cluster centroids as reference points, therefore addressing scalability and computing demands when M>N. We develop a backstepping controller for low-level vehicle management that incorporates sliding-mode components to enhance disturbance rejection and a boundary-layer saturation function to mitigate chattering. The controller is designed for a 12-state quadrotor model and incorporates realistic actuator saturation control. Numerical simulations using M=9 attackers and N=3 defenders illustrate the efficacy of the proposed method: defenders consistently intercept the three cluster centroids in less than 7 seconds, with tracking errors approaching (near) zero in around 3 seconds; control signals are maintained as smooth and devoid of chattering. The work emphasizes the advantages of centroid-based assignment for swarm defense, the resilience of sliding-mode backstepping amongst nonlinear quadrotor dynamics, and the feasibility of centralized sensing and assignment.