Hierarchical Packaged Task Migration for Autonomous Unmanned Swarms in Multiplex Networked Adversarial Scenarios

Abstract

Autonomous swarms use multiple communication, sensing, and service layers, so capability loss can displace tasks and amplify risk through load redistribution. This paper develops a task-migration framework for dependent agent and layer failures. A Clayton copula models within-epoch dependence, and a dimensionless load-propagation matrix characterizes cross-epoch cascades through a spectral-radius stability criterion. We propose Hierarchical Packaged Task Migration (HPTM), which constructs a feasible post-disturbance assignment, migrates cohesive task packages across layers, and refines them through balanced hierarchical decomposition within each layer. A normalized utility combines completion cost with expected completed workload, and every accepted operation uses its exact marginal change. The algorithm preserves feasibility, terminates finitely, and achieves $\epsilon $ -local optimality over its enumerated neighborhoods. Paired experiments show that HPTM improves utility and cost relative to capability-first, risk-aware, and objective-matched auction baselines. Certified small-instance studies place its solutions close to the exhaustive optimum, while sparse screening extends the method to large task sets with a practical quality–runtime trade-off. Note to Practitioners—HPTM supports centralized post-disturbance replanning when task checkpoints, service rates, and migration paths are available. Each layer derives its migration trigger from the current capacity margin, risk budget, and a validated bound on per-agent capability loss. Mission policy or calibration data determine the weights for cost versus completion, skill versus location similarity, and adaptability versus isolation. The direct implementation suits planning epochs that accommodate exhaustive package search, whereas sparse screening lowers latency for larger task sets at a moderate utility cost. Deployments with tighter deadlines can combine screening with incremental risk updates, compiled scoring kernels, or distributed package evaluation. Cases without a feasible checkpoint-to-destination path require task rejection or mission-level replanning.