Cut-in scenario prediction for automated vehicles

Truck platooning is gaining more and more interest thanks to the benefits on improved traffic efficiency, reduced fuel consumption and emissions. To gain these benefits, it typically involves small following distances (0.8 s - 0.3 s). Due to the small following distances, the cut-in manoeuvre of target vehicles becomes safety critical and requires the platooning system to take action as soon as possible. This work shows how machine learning can be used for the prediction of a cut-in manoeuvre of a vehicle, which we refer to as target vehicle, from a host vehicle perspective. A real-life driving experiment was performed to measure several cut-ins that were manually annotated. Measurements are gathered with a lidar installed on the host vehicle and consequently used to train several well-known machine learning algorithms such as Logistic Regression, Random Forest, Support Vector Machine, Adaboost and an Ensemble of the previous models. The Ensemble model achieves the best results. This method is capable of predicting cut-ins prior to their occurrence, with an f 1 score of 62.28 % on the test set. Moreover, over 60% of the cut-ins are correctly predicted more than one second before the corresponding vehicle crosses the lane marker.