A dependable detection mechanism for intersection management of Connected Autonomous Vehicles

Traffic intersections will become automated in the near future with the advent of Connected Autonomous Vehicles (CAVs). Researchers have proposed intersection management approaches that use the position and velocity that are reported by vehicles to compute a schedule for vehicles to safely and efficiently traverse the intersection. However, a vehicle may fail to follow intersection manager (IM) scheduling commands due to erroneous sensor readings or unexpected incidents like engine failure, which can cause an accident if the failure happens inside the intersection. Additionally, rogue vehicles can take the advantage of the IM by providing false position and velocity data and cause traffic congestion. In this paper, we present a new technique and infrastructure to detect anomalies and inform the IM. We propose a vision system that can monitor the position of incoming vehicles and provide real-time data for the IM. The IM can use this data to verify the trajectories of CAVs and broadcast a warning when a vehicle fails to follow commands, making the IM more resilient against attacks and false data. We implemented our method by building infrastructure for an intersection with 1/10 scale model CAVs. Results show our method, when combined with an IM dataflows, is more dependable in the event of a failure compared to an IM without it.