Uncertainty-Gated Distributed Dynamic Task Allocation for Multi-UAV Systems Under Intermittent Communication

Dynamic task allocation in multi-unmanned-aerial-vehicle (multi-UAV) teams must account for uncertain task return, changing reachability, intermittent communication, and terminal-return energy. We develop UCDG-PD for one-task-per-UAV allocation. At each decision epoch, an obstacle-aware bipartite graph keeps only capability-compatible edges satisfying deadline, outbound-route, and task-to-terminal return-energy constraints. A lower-confidence bound converts an edge-level reliability requirement into deterministic admission. On the surviving graph, agents use an idle-augmented entropic assignment, dynamic average tracking of task demand, neighbor-mixed prices, and task-local leases to resolve conflicts without global edge sorting. We derive the chance-constraint reduction and prove uniqueness of the entropic relaxation, characterize consensus fixed points through KKT conditions, bound demand-tracking disagreement under jointly connected mixing, and establish binary lease feasibility. Controlled trajectory-level simulation models UAV motion, energy, task release and expiry, priority shocks, no-fly zones, stochastic links, and terminal returns. Across 28 common-seed trials, UCDG-PD achieves conditional execution success of 0.995, compared with 0.964 for risk-neutral PD and 0.987 for a centralized auction; successful-task coverage is 0.770 versus 0.838 for the auction. A 1600-residual held-out audit gives coverage of 0.944 (95% Wilson interval: 0.932–0.955) against a nominal value of 0.950. These results quantify the reliability–coverage–value trade-off under the defined trajectory-level simulation protocol.