Online Distributed Offloading of Time-Sensitive Vehicular Tasks in Edge-Cloud Systems
Te-Yi Kan, Konstantinos Psounis · IEEE Transactions on Vehicular Technology · 2025
With the integration of several types of sensors, vehicles are able to detect and identify objects in their surrounding, enabling them to provide vital road information to Automated Driving Systems (ADS) or drivers using Advanced Driving Assistance Systems (ADAS) to perform functions such as crash avoidance, crosswalk detection, navigation, etc. Recently, it has been proposed to identify, process, and even display information using machine learning (ML) modules. However, performing ML tasks within vehicles is challenging due to their resource-intensive and delay-sensitive nature. To this end, we investigate an edge-cloud-assisted system and propose a Threshold-based Online Distributed Offloading and Resource Allocation mechanism (TODORA) to reduce task duration by optimizing task offloading decisions and resource allocation. TODORA takes into account communication and other protocol-related overheads, making it applicable to real-world scenarios. Simulation results demonstrate the superiority of TODORA over the periodic optimal solution and state-of-the-art practical schemes. We also show that the computation overhead of TODORA is relatively low as compared to the overall task duration and delay tolerance of tasks, verifying its suitability in practical systems.