Multiagent Reinforcement Learning for Task Offloading of Space/Aerial-Assisted Edge Computing

The task offloading in space-aerial-ground integrated network (SAGIN) has been envisioned as a challenging issue. In this paper, we investigate space/aerial-assisted edge computing architecture considering whether to take advantage of server mounted on the unmanned aerial vehicle and satellite for or not. By optimizing energy consumption completion delay, formulate NP-hard non-convex optimization problem minimize computation cost, limited by capacity availability constraints. formulating Markov decision process (MDP), propose multiagent deep reinforcement learning (MADRL)-based scheme obtain optimal policies dynamic request stochastic time-varying channel conditions, while ensuring quality-of-service requirements. Finally, simulation results demonstrate learned from our proposed algorithm that can substantially reduce average cost compared other three single agent schemes.