Estimated Interception Time Coupling of Task Assignment, Velocity-Predicted Guidance, and Safety Filtering for Multi-UAV Interception

Full abstract

Read the full abstract

Low-altitude incursions by multiple unmanned aerial vehicles (UAVs) threaten maritime critical infrastructure because defenders face numerous intruders, short engagement windows and target speed advantage, and short engagement windows. Conventional distance-based assignment neglects target motion, pursuit of the current target position causes tail chasing, and soft inter-UAV repulsion cannot guarantee separation. This paper proposes a cooperative interception method that couples task assignment, guidance, and safety execution through a shared state snapshot and a common estimated interception time. This coupling coordinates who intercepts each target, where interception is attempted, and how inter-UAV safety corrections are applied. The task layer uses Estimated Interception Time-based Coverage-Constrained Hungarian Assignment (EIT-CCHA) with mobility-aware costs, coverage-priority slots, and termination/infeasibility-triggered reassignment. The guidance layer uses Three-Dimensional Velocity-Predicted Lead Interception Guidance (3D-VPLIG) to combine target-velocity prediction with analytic intercept time when defining the aim point. The safety layer applies a Centralized Three-Dimensional High-Order Control Barrier Function Safety Filter (C3D-HOCBF-SF) formulated as a two-stage lexicographic second-order cone program. Across five unequal-scale and three equal-size scenarios, evaluated as offline algorithmic simulations, 3D-VPLIG achieves scenario-equal-weighted interception rates of 95.21% and 96.68%, and no real-time deployability is claimed at any of the evaluated scales. Its claimed advantage is not dominance over the nine reference guidance laws adapted to the common double-integrator interface but per-target interception efficiency and the shared encounter-time formulation: the assignment cost and the guidance aim point both follow from a constant-velocity encounter-time model, although only guidance is updated at every control step. This is a model-based prediction rather than a finite-time reachability guarantee of the constrained double-integrator kinematics. The strongest adapted baseline, FRPN, remains statistically indistinguishable in protected-zone penetration (1.66% versus 1.74%; paired difference +0.074 percentage points, 95% interval [−0.297,+0.441], p≈0.69). Ablations show that motion-aware assignment adds 1.528 percentage points. The archived collision counts may omit events concurrent with capture and require recomputation before a collision-free safety claim can be made.