Prediction of ego vehicle trajectories based on driver intention and environmental context

By knowing what the driver will do next, it is possible to assist drivers only as much as necessary. Especially in level two safety relevant collision avoidance systems, it is beneficial to know if the driver will do an evasion or a braking maneuver on his own to avoid a collision with a pedestrian or another object in the environment. This work presents an algorithm that outputs the probability of a sporty or comfortable braking or evasion maneuver, a follow lane maneuver or no change in driving behavior at all. Furthermore, the algorithm predicts respective trajectory courses of these discrete maneuvers from the vehicle we are looking at (ego vehicle). By evaluating information about the environmental context, driving data and the current visual and manual driver state in a Bayesian network, the probabilities for ego vehicle trajectories are predicted. The generation of the discrete trajectories is based on the single-track model combined with current driver information, such as reaction time. In order to map drivers' natural behavior in predicting ego vehicle trajectories, a study on driver collision avoidance maneuvers was conducted and analyzed in 2018. In total, 45 subjects performed 1260 braking and 1890 evasion maneuvers, in two different manners (sporty and comfortable). Validation of the algorithm in real world showed reliable prediction of ego vehicle trajectories courses and probabilities in different driver states.