Teaching Vehicles to Anticipate: A Systematic Study on Probabilistic Behavior Prediction Using Large Data Sets

Abstract

By observing their environment as well as other traffic participants, humans are enabled to drive road vehicles safely. Vehicle passengers, however, perceive a notable difference between non-experienced and experienced drivers. In particular, they may get the impression that the latter ones anticipate what will happen in the next few moments and consider these foresights in their driving behavior. To make the driving style of automated vehicles comparable to the one of human drivers with respect to comfort and perceived safety, the aforementioned anticipation skills need to become a built-in feature of self-driving vehicles. This article provides a systematic comparison of methods and strategies to generate this intention for self-driving cars using machine learning techniques. To implement and test these algorithms we use a large data set collected over more than 30 000 km of highway driving and containing approximately 40 000 realworld driving situations. We further show that it is possible to classify driving maneuvers upcoming within the next 5 s with an Area Under the ROC Curve (AUC) above 0.92 for all defined maneuver classes. This enables us to predict the lateral position with a prediction horizon of 5 s with a median lateral error of less than 0.21 m.

Predicting the Time Until a Vehicle Changes the Lane Using LSTM-Based Recurrent Neural Networks

“…In [3], the problem of predicting the future positions of surrounding vehicles is systematically investigated from a machine learning point of view using a non-public data set. Among the considered approaches and techniques, the combination of a multilayer perceptron (MLP) as lane change classifier and three Gaussian mixture regressors as position estimators in a mixture of experts shows the best performance.…”

“…The latter are used to aggregate different position estimates being characteristic for the respective maneuvers. In [4], the approach of [3] has been adopted to the publicly available highD data set [5], showing an improved maneuver classification performance with an area under the receiver operating characteristic curve of over 97 % at a prediction horizon of 5 s. Additionally, [4] studies the impact of external conditions (e. g. traffic density) on the driving behavior as well as on the system's prediction performance.…”

“…Other approaches try to infer the future position or a spatial probability distribution [3], [4], [13]- [16]. As [13] shows, it is promising to perform the position prediction in a divide and conquer manner.…”

“…These probabilities serve as mixture weights in a mixture of experts predicting future positions. This approach has been extended by the abovementioned works [3], [4], which have replaced the random forest by an MLP. The evaluations presented in [4] show a median lateral prediction error of 0.18 m on the highD data set at a prediction horizon of 5 s. A similar strategy is applied by [14].…”

“…The present work builds upon the general approach we described in [3], [4] but follows a fundamentally different idea. We replaced the previously used multilayer perceptron (MLP) for lane change classification by a long short-term memory (LSTM)-based recurrent neural network (RNN) predicting the time to an upcoming lane change.…”