Hybrid Sensor Network with Edge Computing for AI Applications of Connected Vehicles

For supporting Artificial Intelligence (AI) applications of connected vehicles, sensing data is collected by Service Providers (SPs) as input to AI models to execute model inference. Based on the inference results, SPs respond to users' requests. To ensure the quality of service (QoS), enhancing the sensing quality of data collection and shortening the latency of inference execution are two crucial issues. To address these issues, we propose the integration of hybrid sensor network and edge computing. Hybrid sensor network enables the cooperation of dynamic vehicular nodes and static sensor nodes for improving sensing quality. Edge computing fulfills local processing of sensing data in edge servers to improve the overall performance of services. After that, we study the problem for SP-side assigning sensing tasks and corresponding rewards between vehicular nodes and sensor nodes. A three-party Stackelberg game is leveraged to design the task assignment scheme, which allows the three parties to reach a deal with optimal pricing strategies and sensing strategies. We also develop a resource allocation scheme which enables SPs to optimally allocate computation resource of edge servers for minimizing the delay of inference execution. Numerical results indicate that the proposed task assignment scheme based on hybrid sensor network outperforms the schemes based on pure vehicular nodes or sensor nodes. The designed resource allocation scheme achieves the convergence of 4.2 times faster than that of the greedy algorithm.