Federated Learning-Based Travel Time Prediction for Adaptive Congestion Control using 5G-Assisted Roundabout Traffic Networks

As urbanization accelerates and traffic volumes increase, managing congestion in unsignalized roundabouts becomes increasingly complex, particularly in mixed-traffic environments with human-driven vehicles (HVs), connected and automated vehicles (CAVs), and robotic vehicles (RVs). This study proposes a Roundabout Traffic Coordination Mechanism (RTCM), an integrated framework that combines Federated Learning-based Travel Time Prediction (FL–TTP) with hybrid reinforcement learning and model predictive control (RL–MPC). In this design, FL–TTP generates decentralized short-term travel time forecasts that serve as input to the RL–MPC controller within RTCM, where RL adapts long-term policies and MPC enforces short-horizon safety. The system operates on a 5G-assisted vehicle-to-everything (5G-V2X) network, ensuring low-latency synchronization between prediction and control. Simulations at the Nettakallappa Circle roundabout in Bangalore show that FL–TTP reduces waiting time by 26.5%, delay by 28.5%, and RMSE by 18.7% compared with centralized baselines, while the complete RTCM further decreases queue length by 41.2% and improves throughput by 22.3% over standalone RL and MPC. Supported by 5G communication validated using MATLAB 5G Toolbox (R2024b), the framework achieves a reduction in uplink latency from 40 ms to 4.8 ms, downlink latency from 30 ms to 1.7 ms, and a round-trip time of 12.3 ms, while improving packet delivery from 98.5% to 99.999%. These values, benchmarked against the 3GPP Ultra-Reliable Low-Latency Communication (URLLC)/V2X specifications (TS 22.261, TS 38.300, TR 22.886), confirm that the proposed RTCM is both numerically effective and compliant with 5G communication standards. The results establish a scalable, privacy-preserving, and communication-efficient framework for real-time traffic control in unsignalized roundabouts.