Risk-Diffusion Auction for Spatially Coupled Multi-UAV Functional Task Allocation at the Edge

Abstract

Edge computing allows a nearby server or leader UAV to plan with current sensor data. In many multi-UAV missions, mandatory tasks create mission value, while functional tasks such as reconnaissance and jamming reduce the risk of those mandatory tasks. A functional task is useful only when it is placed close enough to the tasks it supports, but executing it also adds travel, payload, and exposure costs. Task selection and route planning are therefore coupled through space. Many existing allocators score tasks separately and do not represent this coupling directly. We propose FTRDA, a Functional-Task Risk-Diffusion Allocation algorithm. FTRDA builds a battlefield dependency graph (BDG) with distance-decayed support edges. An exponential model converts the total support received by a task into its effective risk. The algorithm then ranks functional tasks by risk reduction per unit cost, removes redundant supporters, and uses a heterogeneous sequential single-item auction to build feasible UAV routes. The auction runs on a central edge node and enforces UAV type, capacity, and range constraints. Experiments on $50\times 50$ grid scenarios vary fleet size, task mix, risk, and support radius. FTRDA reduces mission cost by 8% to 30% relative to six heuristic baselines. On small and medium instances, its solution quality remains close to the CPLEX reference while requiring much less time; on the largest reported comparison, the runtime difference is about four orders of magnitude. The evaluation also reports statistical tests, planning latency, message volume, and edge memory separately. Under the stated static and full-observability assumptions, FTRDA provides a lightweight planning method for IoT-connected UAV fleets. The edge planner is centralized; the method is not a fully distributed swarm-intelligence protocol.