Federated Learning on the Road Autonomous Controller Design for Connected and Autonomous Vehicles

The deployment of future intelligent transportation systems is contingent upon seamless and reliable operation connected autonomous vehicles (CAVs). One key challenge in developing CAVs the design an controller that can accurately execute near real-time control decisions, such as a quick acceleration when merging to highway frequent speed changes stop-and-go traffic. However, use conventional feedback controllers or traditional learning-based controllers, solely trained by each CAV’s local data, cannot guarantee robust performance over wide range road conditions traffic dynamics. In this paper, new federated learning (FL) framework enabled large-scale wireless connectivity proposed for designing CAVs. framework, models used are collaboratively among group To capture varying CAV participation FL training process diverse data quality CAVs, novel dynamic proximal (DFP) algorithm accounts mobility fading channels, well unbalanced non-independent identically distributed across A rigorous convergence analysis performed identify how fast converge using optimal controller. particular, impacts on DFP explicitly analyzed. Leveraging analysis, incentive mechanism based contract theory designed improve speed. Simulation results real vehicular traces show DFP-based track target time under different scenarios. Moreover, has much faster compared popular algorithms averaging (FedAvg) (FedProx). also validate feasibility contract-theoretic 40% baselines.