D2D-Assisted MEC Bidirectional Task Offloading and Resource Allocation in IIoT

Mobile Edge Computing (MEC) and Device-to-Device (D2D) communication show remarkable effects in solving the problem of insufficient device computing resources in Industrial Internet of Things (IIoT). Tasks can be offloaded to MEC servers or idle robots to reduce computation delay and avoid network congestion. However, with D2D communication, the dimensionality explosion problem arises due to complex offloading strategies in real-world scenarios. To address this concern, we design a Bidirectional Task Offloading and Resource Allocation (BTORA) algorithm based on multi-agent deep reinforcement learning. The BTORA algorithm employs an action mask mechanism to dynamically limit the offloading action space of the task generating device to fully utilize the computational resources. In addition, Transformer is utilized to allow Mixing network to focus on the key informations of task offloading, thus optimizing the task offloading decision. Simulation results show that the BTORA algorithm outperforms existing algorithms in terms of cumulative reward, average processing delay, and average offloading success rate while maintaining a lightweight model.