Publication

This project focuses on enhancing the safety of connected and autonomous vehicles (CAVs) operating within complex mixed traffic environments, where both human-driven vehicles (HDVs) and CAVs coexist. In such scenarios, CAVs are subject to multiple sources of uncertainty, including unpredictable human behavior, incomplete or delayed communication, and potential cyber threats. To address these challenges in a systematic and comprehensive way, the project concentrated on two principal objectives. The first was to develop models and techniques for perceiving and quantifying uncertainties that arise in mixed traffic. The second was to establish adaptive and robust operational control strategies for CAVs that can enhance safety under uncertain conditions. In pursuit of the first objective, the project developed advanced models to capture and quantify the variability inherent in human driving behavior and environmental complexity. A probabilistic trajectory prediction framework was proposed to incorporate behavioral heterogeneity among human drivers. By learning from historical trajectory data and extracting driver-specific behavior features, the model significantly improves the accuracy of future movement predictions. This contributes to the anticipation of potentially dangerous interactions and allows CAVs to plan more cautiously in uncertain surroundings. In addition, a vision-based traffic risk estimation framework was established to quantify human-perceived risk using features extracted from traffic scenes. The system was trained using human-labeled risk scores and demonstrates the ability to distinguish between scenes of varying perceived safety. These models enable real-time identification of high-risk scenarios and form a foundation for risk-aware CAV operation. The project also addressed communication-related uncertainty, which arises from issues such as packet loss and delayed information exchange in V2V communication systems. This type of uncertainty can lead to discontinuities in reference trajectories and provoke unstable responses in CAV control systems. To mitigate the resulting transients and suppress instability, a trajectory smoothing method was developed based on a variant sigmoid function. This strategy reduces the demand on communication bandwidth and ensures that trajectory tracking remains smooth and safe, even under degraded communication conditions. These contributions jointly strengthen the perception and quantification of uncertainties affecting CAV safety in real-world driving environments. To meet the second objective, the project delivered a set of methods that emphasize adaptive planning and control under uncertainty, and validated them through simulation-based experimentation. A trajectory planning and signal coordination strategy was
developed for CAVs navigating through dynamic urban intersections. This method incorporates vehicle-level dynamics into intersection-level decision-making to manage uncertainty in traffic flow and ensure both stability and safety. It enables CAVs to adjust their trajectories in response to evolving traffic conditions and promotes coordinated behavior in mixed traffic streams. In addition to planning, the project also proposed a control strategy tailored to bidirectional platoons of CAVs. This adaptive control method explicitly accounts for challenges such as actuator saturation and discontinuous trajectory inputs caused by communication failure. The control framework incorporates both internal uncertainties, including model parameter variability, and external uncertainties, such as unpredictable HDV maneuvers or information delays. By using an adaptive mechanism to adjust control inputs in real time, the method ensures that CAVs can maintain tracking performance and inter-vehicle stability despite uncertain and evolving operational conditions. All developed methods were implemented and evaluated in a simulation
environment built on the CarSim platform. This environment provided a realistic and controllable setting to validate control and planning strategies under a wide range of traffic conditions, behavioral variations, and uncertainty levels. The simulation results confirmed the effectiveness of the proposed methods in improving safety, stability, and responsiveness of CAV operations in heterogeneous traffic environments. The research outcomes contribute to the advancement of safety-aware autonomous driving technologies and lay a strong foundation for future experimental validation and large-scale deployment of CAV systems in real-world scenarios.