Deep reinforcement learning-based dynamical task offloading for mobile edge computing

Edge computing holds considerable potential to augment the efficiency of computational and energy resources in mobile edge computing (MEC) through the application of task offloading technology. Task offloading involves the delegation of multiple tasks, generated by devices with limited resources, to an edge server or other devices equipped with enhanced capabilities, which facilitates more efficient processing times and energy utilization. This paper addresses significant challenges in MEC environments, particularly the issue of inaccurately perceived task rewards due to queuing delays in task offloading, which adversely affects the performance of deep reinforcement learning (DRL) strategies. We introduce the reward-aware proximal policy optimization (RAPPO) algorithm, which is specifically designed to enhance the accuracy of reward perception by refining data sampling methods. This adjustment enables RAPPO to accurately associate rewards with their respective tasks despite inherent delays, thereby enhancing decision-making and improving overall system performance in dynamic and uncertain environments. Simulation results demonstrate that the proposed algorithm achieves competitive performance compared to existing baseline schemes.