Multi-Agent Deep Reinforcement Learning-Based Decentralized Computation Offloading in Mobile Edge Computing

It is hardly possible to support latency-sensitive and computational-intensive applications for mobile devices with limited battery capacity and low computing resources. The development of mobile edge computing and wireless power transfer technologies enable mobile devices to offload computing tasks to edge servers and harvest energy to extend their battery lifetime. However, computation offloading faces challenges such as the limited computing resources of the edge server, the quality of the available communication channel, and the limited time for energy harvesting. In this paper, we study the joint problem of decentralized computation offloading and resource allocation in the dynamic environment of mobile edge computing. To this end, we propose a multi-agent deep reinforcement learning-based offloading scheme that considers the cooperation between mobile devices to adjust their strategies. To be specific, we propose an improved version of the multi-agent deep deterministic policy gradient algorithm by employing the features of clipped double Q-learning, delayed policy update, target policy smoothing, and prioritized experience replay. The simulation results reveal that the proposed offloading scheme has better convergence performance than other baseline methods and also reduces the average energy consumption, average processing delay and task failure rate.