Fast Role-Switching-Aware Task Offloading for Cloud-Edge-End Collaboration in Space-Air-Ground Integrated Networks

Cloud-edge-end task offloading in space-air-ground integrated networks is complicated by two coupled dynamics: rapidly changing low-Earth-orbit satellite coverage and energy-dependent UAV functionality. Existing node-selection models do not directly represent multi-hop relay and cloud-forward paths, and repeated cold-start optimization wastes time when adjacent slots remain partly correlated. We formulate a causal path-role-resource model in which each task selects one complete service path or an explicit drop path, and every collaborative node takes one mutually exclusive role from compute, relay, cloud-forward, and sleep. LEO feasibility follows orbit-driven current visibility and predicted coverage; UAV energy separates hovering, flight, communication, computation, and switching. RS-WBCD combines feasibility repair, block updates, and monotone acceptance. Across 100 slots and 20 matched seeds, it completed 93.33% of tasks and required 1.199 iterations and 4.223 ms per slot. Relative to the no-warm variant, warm starting reduced iterations by 16.2% and runtime by 14.4%. These results support faster causal adaptation under the tested orbit approximation, without a claim of global optimality.