Dynamic Object Map Based Architecture for Robust CVS Systems

Connected and Autonomous Vehicles (CAV) rely on information obtained from sensors and communication to make decisions. In a Cooperative Vehicle Safety (CVS) system, information from remote vehicles (RV) is available at the host vehicle (HV) through the wireless network. Safety applications such as crash warning algorithms use this information to estimate the RV and HV states. However, this information is uncertain and sparse due to communication losses, limitations of communication protocols in high congestion scenarios, and perception errors caused by sensor limitations. In this paper we present a novel approach to improve the robustness of the CVS systems, by proposing an architecture that divide application and information/perception subsystems and a novel prediction method based on non-parametric Bayesian inference to mitigate the detrimental effect of data loss on the performance of safety applications. The architecture is validated with simulations and in a real environment using a remote vehicle emulator (RVE) based on a Denso OBU, which allows the joint study of the CVS applications and its underlying communication system. We study the impact of using different vehicle tracking (prediction) techniques and demonstrate the performance improvement potential of this approach.