This paper examines how federated learning (FL) can be made more reliable and efficient in vehicular environments, where vehicles act as learning clients while continuously moving across the road network. Mobility creates challenges such as fluctuating wireless connectivity, variable computation capacity, and unpredictable client availability, all of which can degrade FL performance. To address these issues, the authors propose VREM-FL, a mobility-aware framework that jointly considers computation load and scheduling decisions to improve the quality and timeliness of model updates. A key contribution of the work is the introduction of a Vehicular Resource Evaluation Metric (VREM), which quantifies how suitable a vehicle is for participating in FL rounds based on factors such as expected contact duration, computation capability, and communication reliability. By incorporating mobility predictions into the scheduling process, the server can prioritize vehicles that are more likely to complete their local training and successfully upload updates before disconnection.
The study also presents a co-design strategy that optimizes both which vehicles are selected and how much computation they should perform, balancing learning accuracy with system constraints. Through analytical modeling and simulation, the authors assess how VREM-FL improves model convergence under realistic vehicular mobility patterns compared to baseline scheduling approaches. Overall, the paper highlights the importance of mobility-aware resource management in enabling scalable and dependable FL for intelligent transportation systems. By aligning computation and scheduling with the movement behavior of vehicles, VREM-FL provides a more robust foundation for future connected and autonomous driving applications.
VREM-FL_Mobility-Aware_Computation-Scheduling_Co-Design_for_Vehicular_Federated_Learning (1)