Traffic Control based on CARMA Platform for Maximal Traffic Mobility and Safety

Abstract

This project developed and evaluated an online adaptive platoon control framework for connected and automated vehicles (CAVs) to improve mobility and safety in dynamic traffic environments. The proposed Physics Enhanced Residual Learning (PERL) framework integrates a physics-based centralized controller, which models vehicle dynamics to ensure stability and interpretability, with a neural network residual module that adaptively corrects unmodeled dynamics and disturbances in real time. Our work can contribute to the U.S. DOT CARMA Platform by serving as a tactical-level cooperative longitudinal control component within the platooning plugin. In this role, PERL has the potential to enhance CARMA’s mobility operations layer by enabling real-time gap regulation, adaptive disturbance mitigation, and improved string stability under heterogeneous and mixed traffic scenarios. By providing a flexible yet interpretable control approach, PERL could support CARMA’s cooperative driving automation objectives and facilitate smoother integration of platooning strategies into broader transportation system management and operations. High-fidelity simulations and scaled platform experiments demonstrated that PERL outperforms both pure physics-based and pure learning-based controllers, reducing cumulative position and speed errors by more than 50% in simulation and up to 99% in scaled platform tests. These results suggest that the proposed framework could strengthen CARMA’s cooperative driving capabilities, improve the resilience of platoon control, and support the deployment of safe, efficient, and adaptive CAV platooning in real-world operations.