Robust Deep Reinforcement Learning for Task Scheduling in MEC-Enabled IoT Networks

Abstract

Robust Reinforcement Learning (RL) based task scheduling approaches can address the inherent tradeoff between energy consumption and deadline violation in a Multi-access Edge Computing (MEC) based Internet of Things (IoT) network, while maintaining robustness against changes in the task arrival rate. However, tabular robust RL algorithms suffer from high computational and storage complexity, and therefore are not scalable to a system with a large number of IoT nodes. To this end, in this paper, we propose a robust deep RL based task scheduling algorithm to solve the underlying Robust-Return Constrained Markov Decision Process (R2CMDP) problem. The proposed algorithm introduces a tunable amount of robustness in the solution of the RL framework. Complexity analysis and ns-3 simulation results are presented to demonstrate the efficacy of our algorithm.