Abstract
Connected and Automated Vehicles (CAVs) can substantially improve traffic efficiency, safety, and energy sustainability through cooperative platooning. However, under early deployment conditions with low market penetration rates (MPRs), the benefits of platooning are significantly constrained by heterogeneous human driving behavior, partial vehicle connectivity, and limited coordination strategies in mixed traffic. This dissertation addresses these challenges through a comprehensive framework that integrates human-like behavioral modeling, connectivity-enabled vehicle identification, cooperative control, and multi-agent decision-making, validated through high-fidelity simulation and field experiments.
To improve the realism of mixed-traffic modeling, a human-like extension of the Intelligent Driver Model is developed. By incorporating stochastic and adaptive headway dynamics calibrated from real-world trajectory data, the proposed model captures heterogeneous driver behavior more accurately than conventional fixed-parameter models. Results show reductions of approximately 14–22% in speed and acceleration errors compared with standard IDM formulations, providing a more reliable foundation for evaluating cooperative platooning strategies.
To address partial connectivity challenges, a Connected Vehicle Identification System (CVIS) is developed by integrating onboard sensing with vehicle-to-everything (V2X) communication. The proposed system identifies both immediate and upstream connected vehicles beyond line-of-sight limitations, enabling robust cooperative partner detection in partially connected traffic streams. Field experiments using a hardware prototype demonstrate near-perfect identification accuracy under realistic sensing and communication conditions.
For urban arterial environments with signalized intersections and frequent stop-and-go traffic, connectivity-aware cooperative longitudinal control strategies are developed. Linear feedback and Model Predictive Control (MPC) approaches, combined with signal phase and timing (SPaT)-based speed trajectory optimization, significantly improve traffic flow stability. Experimental and simulation results indicate reductions of up to 55% in spacing error, approximately 11% in acceleration variability, and about 13% in fuel consumption, while effectively mitigating stop-and-go oscillations.
For highway scenarios with low CAV penetration, cooperative lane-change decision-making is developed to facilitate platoon formation. A CNN-QMIX multi-agent reinforcement learning framework enables coordinated decisions among multiple CAVs under dynamic traffic conditions. Compared with rule-based approaches such as MOBIL and greedy strategies, the proposed method increases platoon formation rates by approximately 10%, while improving average traffic speed and energy efficiency across varying MPR levels.
Comprehensive evaluations across multiple traffic scenarios and penetration levels demonstrate that the proposed integrated framework significantly enhances system-level performance by improving traffic efficiency, reducing energy consumption, and stabilizing traffic flow. By integrating behavioral realism, connectivity-aware perception, control optimization, and multi-agent learning, this dissertation provides a scalable pathway toward real-world deployment of cooperative automated driving systems.
