Section: New Results

Decision making for automated vehicles in urban environments

Participants : Pierre de Beaucorps, Thomas Streubel, Anne Verroust-Blondet, Fawzi Nashashibi.

The development of automated vehicles in urban environments requires a robust sensing system followed by an adaptive situation assessment. This is the basis for smart decision making in the driving process without collisions or taking high risks. We address this aspect of automated driving in a project with the sensor developer VALEO. The focus is on complex urban traffic scenarios, e.g. intersections and roundabouts, including multiple road users.

In a first step, we developed a new multi-agent driving simulation as a tool to explore human behavior in relevant traffic scenarios. We conducted a study with 10 test persons driving in a scene with one dummy car to acquire data and understand the human decision process in risky situations. This data was used to retrieving speed profiles for the trajectory planning. The path planning was established with Bezier curves. Further, a robust decision making algorithm utilizes the trajectory planning coupled with a risk assessment. The latter is estimating the post-encroachment time (PET), which is the time between one vehicle leaving a collision zone in an intersection area and the other car entering this same zone. Based on this estimation a risk is assigned to every predetermined speed profile and the one with lowest acceptable risk is chosen to be send to the controller of the automated vehicle. The results showed better performance than the drivers in our study. The so equipped automated vehicle is integrated in our simulation environment and was presented to our project partners in several intersection and roundabout scenarios with a real driver in the same scene.