Mobility and Reliability-Aware Task Offloading in Dynamic MEC-IoT Networks using Primal-Dual RL

Abstract

Dynamic task offloading in Multi-access Edge Computing (MEC)-enabled IoT networks is challenging due to timevarying wireless channels, mobility, limited device energy, and unreliable MEC service conditions. Existing works mainly optimize delay and energy, but many of them do not jointly consider mobility-driven connectivity changes, and MEC reliability in a unified online decision framework. To address this gap, this paper proposes a mobility and reliability-aware task offloading scheme for dynamic MEC-IoT networks. The problem is formulated as a constrained Markov decision process (CMDP) under deadline, device-energy, reliability, and MEC queue constraints. To solve it online, a primal-dual actor-critic framework with masked action selection is developed. Simulation results show that the proposed scheme improves the deadline satisfaction ratio by up to 12.2% and reduces the normalized cost by up to 19.2% compared with benchmark methods. The proposed approach is further validated through real-world testbed experiments to demonstrate its practical feasibility.