The field of edge computing in space-air-ground integrated networks is rapidly advancing, with a focus on developing innovative solutions to address the challenges of coordinating decisions across heterogeneous nodes, modeling complex factors, and handling real-time decision-making. Researchers are exploring new architectures, optimization techniques, and machine learning algorithms to improve the efficiency and scalability of edge computing systems. Notably, the use of unmanned aerial vehicles (UAVs) and tethered UAVs is being investigated to enhance the throughput and connectivity of cell-edge ground user equipments.

Recent studies have demonstrated the effectiveness of federated graph-enhanced multi-agent reinforcement learning, distributionally robust optimization, and two-timescale optimization frameworks in improving the performance of edge computing systems. These approaches have shown significant reductions in energy consumption, task completion delay, and communication overhead, while achieving superior convergence stability and scalability.

Some noteworthy papers in this area include: The paper proposing the MADDPG-COCG algorithm, which significantly enhances user-centric performances in terms of aggregated UD cost, task completion delay, and UD energy consumption. The AirFed framework, which achieves a 42.9% reduction in weighted cost compared to state-of-the-art baselines and attains over 99% deadline satisfaction and 94.2% IoT device coverage rate. The two-timescale optimization framework for IAB-enabled heterogeneous UAV networks, which obtains up to 12.2% average throughput gain compared to the MADDPG scheduling method.