Safe Intersection and Merging Coordination of Connected and Automated Vehicles

QC 20230425

Abstract

Connected and automated vehicles (CAVs) are a transformative technology that promises to bring innovative solutions to transportation systems. One of their significant advantages is the elimination of human factors, which makes them capable of resolving the congestion problem prevalent in areas such as ramp merging points and road intersections. Through the sharing of information between vehicles and with the infrastructure, CAVs can coordinate their cross-time cooperatively. This collaborative approach results in avoiding unnecessary stops, which leads to more efficient utilization of the road infrastructure and improved safety for all road participants.

In this licentiate thesis work, we study the problem of formulating coordination strategies for CAVs to efficiently traverse through conflicting regions while maintaining safety with other road participants such as other CAVs or platoons of CAVs or human-driven vehicles (HDVs). We emphasize the challenges when platoons or HDVs are involved and develop coordination solutions on various scales accordingly.   

We start by considering a highway ramp merging scenario that involves platoons of CAVs. In such scenarios, vehicles in a platoon drive in close proximity and create moving barriers for merging traffic. To address this challenge, we propose a bi-level coordination framework. A central coordinator schedules the merging time and speed for all CAVs by solving a mixed integer linear programming (MILP) problem, optimizing traffic performance while maintaining platoon formation. This assigned schedule is executed by each individual vehicle at the control level. When integrated, this framework allows platoons to split occasionally for the merging vehicles, balancing traffic throughput with platoon formation.

In intersections where mixed traffic is present, CAVs need to anticipate the driving behavior of HDVs accurately to plan a safe future trajectory. Due to the unpredictable nature of human behavior, we propose an invariant safe model predictive control (MPC) that considers worst-case scenarios using forward reachable sets to guarantee safety at all times. In addition, to compensate for conservatism, we apply a contingency MPC (CMPC) framework with parallel horizons. One horizon is for safety guarantees in contingency, while the other is for optimizing performance. The methods are applied to general intersection problems through distributed implementation and produce safe and efficient coordination results.

Lastly, we examine the challenges that arise in the real-life implementation of the proposed coordination strategies. To account for potential occlusions, estimation errors, and communication defects, we propose a communication-based and integrated framework from estimation to control. Through set estimation and reachability analysis, we generate robust set estimations of surrounding vehicles and integrate this information with the coordination stack to ensure safe navigation through intersections in an experimental setting.

urn.kb.se/resolve?urn=urn:nbn:se:kth:diva-326014