Probabilistic Collision Threat Assessment for Autonomous Driving at Unsignalized T-Junctions: Merging into Traffic on the Major Road and Being Merged by Traffic on the Minor Road

Abstract

In this paper, we present a probabilistic collision threat assessment algorithm for autonomous driving at unsignalized T-junctions that assesses a given traffic situation at an unsignalized T-junction reliably and robustly for an autonomous vehicle to pass through the unsignalized T-junction safely. To this end, the presented algorithm employs a detailed digital map to predict future paths of observed vehicles and then uses the predicted future paths to identify potential threats—observed vehicles that have the potential to pose a threat to the autonomous vehicle. Next, it establishes vehicle-to-vehicle collision relations with the potential threats. It then employs Bayesian networks and time window filtering to assess the potential threats reliably and robustly regarding the possibility of collision, even under uncertain and incomplete noise data. We have tested and evaluated the presented algorithm through in-vehicle testing at unsignalized T-junctions regarding two representative maneuvers: merging into traffic on the major road and being merged by traffic on the minor road. In-vehicle testing results show that the performance of the presented algorithm is sufficiently reliable to be used in decision-making for autonomous driving at unsignalized T-junctions, in terms of reliability and robustness.