IATS: Information-age aware task scheduling for vehicle-road-cloud cooperative systems

As the Internet of Things (IoT) continues to evolve at an unprecedented pace, smart vehicles can now support various real-time applications like object detection that require timely situational awareness. The freshness of state information is critical for these time-sensitive applications, as it directly affects the timeliness and accuracy of situational awareness. Fresh state information enables smart vehicles to make correct decisions in dynamic environments. However, this aspect is often ignored in prior work. Besides, these applications are typically computation-intensive, posing a challenge to resource-limited smart vehicles. Considering that the state information’s freshness can be characterized using the age of information (AoI) and the vehicle-road-cloud computing architecture is effective in integrating the resources of roadside units and the cloud to assist with processing tasks for vehicles, to minimize the system’s long-term average AoI without violating delay and energy constraints, this paper explores the AoI-aware task scheduling problem in a vehicle-road-cloud cooperative system. To achieve this goal, we first develop an AoI model tailored for smart vehicles within the cooperative system and formulate an AoI optimization problem. In order to tackle the proposed issue, we design a multi-agent reinforcement learning-based task scheduling method that can perform task scheduling in complex, dynamic, and decentralized decision-making environments. The algorithm iteratively trains the network continuously such that all agents obtain the optimal scheduling strategy. Finally, we implement extensive simulations and testbed-based experiments to validate our method. The results indicate that our method reduces the average AoI by 81.91% on average and 95.23% at the highest compared to benchmarking approaches.