An Infrastructure-based Cooperative Driving Framework for Connected and Automated Vehicles

Abstract

Trajectory planning is a key component of the Connected and Automated Vehicle (CAV) autonomy stack. It is a challenging task to plan a trajectory for a CAV that ensures safety, mobility, and comfort, especially in complex urban scenarios. In urban driving scenarios, CAVs need to interact with different road users and follow traffic rules (e.g. turning at traffic signals, yielding at a roundabout), which increases the complexity of the trajectory planning.

One potential solution to help address this challenge is to deploy smart infrastructure, which can enhance microscopic situational awareness to support the trajectory planning of CAVs. An efficient cooperative scheme between the smart infrastructure and CAVs will not only enhance the safety and mobility performance of CAVs but also improve the overall system efficiency. Therefore, in this dissertation, a cooperative planning framework is proposed that given the past trajectories of the detected vehicles, the smart infrastructure manages to provide guidance or warning to CAVs with different applications to assist their trajectory planning.

First, an integrated control framework is proposed to optimize the traffic signals in an urban arterial and provide guidance for the trajectory planning of CAVs in a mixed traffic composition of CAVs, Connected Vehicles (CVs), and Regular Vehicles (RVs). Existing studies suffer from limitations such as assuming 100% penetration rate of the CAV, centralized formulation, or limited at an isolated intersection. Thus, a combination of centralized and decentralized integrated control framework is proposed that the smart infrastructure only provides centralized trajectory planning guidance to the CAVs, and then the CAVs will plan their detailed trajectories individually. The framework is designed for a traffic corridor under mixed traffic conditions.

Second, an anomaly detection model using learning from demonstration is proposed to identify abnormal trajectories when the localization module of a ac{CV} or ac{AV} is under cyber attacks (e.g. ac{GPS} spoofing, sensor attacks). Most cyber defense methods in the literature targeting GPS spoofing attacks or remote sensor attacks require access to the physical signal receivers and may not be readily available in the real-world driving environment. Instead of investigating physical GPS or LiDAR signals, this work proposes a cyber defense model that leverages domain knowledge of transportation and vehicle engineering.

Lastly, a hierarchical Principle Other Vehicle (POV) behavior prediction framework incorporating traffic signal information is proposed to predict vehicle behaviors in urban scenarios with interactive agents. This framework includes a discrete intention prediction module and a continuous trajectory prediction module, and a mixture of offline learning and online prediction strategies are adopted to capture the difference among human drivers. Game theory is utilized to model the interaction between agents explicitly.

In summary, this thesis proposes an infrastructure-based cooperative driving framework to provide a variety of guidance or warnings to CAVs. The framework is validated in the realistic simulation or with real-world datasets. It is expected that the cooperative driving framework can be implemented in the real world to assist the trajectory planning of the CAVs, benefiting their deployment.