Computation Offloading and Resource Allocation in Vehicular MEC: A Parameterized Deep Reinforcement Learning Approach

Rapid advancement of the Internet of vehicles (IoV) technology has created a lot of intelligent vehicle applications, such as real-time navigation, collision avoidance, and autonomous driving, etc., which in turn generates a significant amount of computation-intensive tasks. Due to limited computation capabilities of in-vehicle devices, mobile edge and cloud computing are viewed as promising solutions to offload latency-sensitive tasks to mobile edge computing servers to meet the requirements for euRLLC in 5 G/6 G applications. Vehicular edge computing systems require effective offloading decision-making and resource management to maximize resource utilization. However, existing scheduling schemes suffer a high complexity to deal with multi-dimensional resource optimization problems, thereby introducing a significant latency and compromising resource utilization efficiency. This paper aims to address the issues on latency, energy consumption, and cost jointly within a vehicular edge computing system. In particular, a multi-objective optimization problem is formulated for its joint optimization on computation offloading and resource allocation with the help of a deep reinforcement learning (DRL) approach. To this end, a parameterized deep Q-networks (P-DQN) algorithm is introduced for solving a binary offloading problem. Extensive numerical simulations validate the superiority of the proposed scheme compared to well-known algorithms in the literatures.