Stochastic Driver Velocity Prediction with Environmental Features on Naturalistic Driving Data

Predictive energy management and driver assistance systems are strongly depending on the vehicular velocity forecast including the dynamic behavior of the driver. In this brief two stochastic frameworks are derived to predict the driver depended velocity on basis of environmental features. First, the stochastic model assumes linearity and Gaussian distribution for underlying densities that lead to the Kalman Filter and Rauch-Tung-Striebel Smoother algorithms. Using non-parametric distributions for the second approach which makes numerical methods like particlebased algorithms essential, achieves more accurate prediction results for modeling driver individual behavior. The algorithms are evaluated on a study with eight drivers. Furthermore, the framework is usable for various automotive applications like prediction of lateral vehicle dynamics.