Distributed Energy-efficient Computation Offloading and Trajectory Planning in Aerial Edge Networks

In this paper, we investigate energy-efficient computation offloading and trajectory planning for aerial edge networks, wherein multiple Unmanned Aerial Vehicles (UAVs) cooperate with each other to provide computing service to the ground users. In particular, we formulate the joint optimization of computation offloading, channel allocation, power control, resource allocation, and trajectory planning as a multi-agent deep reinforcement learning (MA-DRL) problem to minimize the system energy consumption while guaranteeing the latency re-quirements of computation tasks. Then, we propose a distributed algorithm to enable UAV s to independently make action decisions based on their local observations, still collaboratively learn their policies for system performance. Numerical results demonstrate that our proposed algorithm can effectively plan the trajectory, reduce the system energy consumption, and improve the latency requirements satisfaction ratio.