Deep Reinforcement Learning for Multi-Hop Offloading in UAV-Assisted Edge Computing

In this article, we propose a unmanned aerial vehicle (UAV)assisted multi-hop edge computing (UAV-assisted MEC) system in which a UE can offload its task to multiple UAVs in a multi-hop fashion. In particular, the UE offloads a task to its nearby UAV, and this UAV can execute a part of the received task and offload the remaining part to its neighboring UAV. The offloading process continues until the task execution is finished. The benefit of this multihop offloading is that the task execution can be finished faster, and the computing load can be shared among multiple UAVs, thus avoiding overloading and congestion. Each node, i.e., the UE or the UAV, needs to determine the task size for offloading to minimize the cumulative energy consumption and latency over the nodes. We formulate a stochastic optimization problem under the dynamics and uncertainty of the UAV-assistedMECsystem. Then, we propose a deep reinforcement learning (DRL) algorithm to solve this problem. Simulation results are provided to demonstrate the effectiveness of the DRL algorithm.