A risk-aware improved grey wolf optimizer algorithm for multi-UAV task allocation in spatially constrained environments

This paper investigates risk-aware task allocation for heterogeneous multi-UAV systems under coupled communication, time-window, workload, spatial-risk, payload, and endurance constraints. Existing formulations may exclude useful traversable routes or underestimate executable service by treating hazardous regions as hard barriers or using all-or-nothing feasibility evaluation. To address these limitations, a risk-aware improved Grey Wolf Optimizer (RA-IGWO) is developed within a partial-completion framework. Risk-prone and communication-degraded regions are modeled as traversable soft-cost areas, while physical obstacles are explicitly penalized and additionally examined under hard-obstacle enforcement. RA-IGWO integrates capacity-aware initialization, opposition-based learning, decoded-space local refinement, and accept-only-improved population updates. The quadratic schedule is retained only as an exploration policy rather than an independently beneficial component. All algorithms are evaluated under the same global number of fitness evaluations budget, with every candidate evaluation counted toward that budget. Experiments on independently generated instances show that RA-IGWO achieves lower objective values than five population-based baselines and remains competitive with a recent discrete adaptive large neighborhood search benchmark method. Robustness analyses show that the conclusions are not tied to a single decoder, obstacle treatment, or objective setting.