PPO-Based Task Offloading With EKF for Position Prediction in RSU-Assisted IoV

In the Internet of vehicles (IoV) environment, roadside units (RSUs) play an important role in data processing because of their powerful environment awareness. However, RSU-to-everything tasks often have stringent low-delay requirements and computing resources of RSUs are limited, which makes it difficult for RSUs to handle all generated computing tasks. In order to meet delay constraints of tasks, a source RSU needs to assign some computing tasks to other network nodes for processing, such as other RSUs or vehicles. In addition, both RSUs and vehicles have limited energy resources. Mobile vehicles exhibit dynamic trajectories. It leads to frequent changes in the set of vehicles within RSU communication ranges. However, GPS positioning errors can affect the accuracy of association of RSUs and vehicles. To address the challenge, this paper utilizes Extended Kalman Filter (EKF) by fusing GPS-based measurement trajectories. In addition, on the one hand, considering that vehicles may leave the communication range of the source RSU during task transmission to a vehicle, leading to task transmission failures, it is necessary to identify which vehicles remain within the RSU communication range during the task transmission period; on the other hand, vehicles may no longer be within the communication range of the source RSU after completing the task processing. Thus, it is necessary to estimate the vehicle position in order to send the task results back to the source RSU by multi-hop transmissions of other RSUs. A weighted averaging method is employed to predict vehicle trajectories according to the historical EKF trajectories. Finally, the task offloading problem after predicting vehicle trajectories is transformed into a 0-1 integer programming problem and further into an Markov decision process (MDP), which is solved by the proximal policy optimization (PPO) algorithm. Extensive experiments validate the effectiveness of the proposed scheme.