作者
Duanjiao Li,Yongchao Liang,Wenxing Sun,Zibin Zhu,Jianguo Zhang
摘要
Multi-robot systems in intelligent warehousing, industrial inspection, disaster relief, and UAV collaboration face complex tasks including target detection, map fusion, path planning, status diagnosis, and log uploading. Limited onboard computing, heterogeneous edge resources, and fluctuating links expose trajectories, maps, images, and status data during offloading, and traditional latency- or energy-only optimization fails to address scalability, security, and privacy together. This study proposes SPP-ESO, a security- and privacy-aware edge offloading strategy based on a three-layer robot-edge-cloud architecture, integrating task profiling, privacy level identification, dynamic trust assessment, candidate node filtering, multi-objective reinforcement learning, differential privacy perturbation, data minimization, and edge-failure migration into a unified model optimizing latency, energy, load balancing, security risk, privacy risk, and failure penalty together. With 200 robots, SPP-ESO achieves 218 ms average latency, 4.73 J energy consumption, 96.4% task completion, 3.6% SLA violation, 3.1% attack success, and 0.22 privacy risk, outperforming local-only, cloud-only, greedy-latency, DRL-offloading, and trust-aware baselines. At 300 robots under high concurrency, completion rate remains ∼94%.