This content is not included in your SAE MOBILUS subscription, or you are not logged in.
Driver’s Response Prediction Using Naturalistic Data Set
ISSN: 2641-9637, e-ISSN: 2641-9645
Published April 02, 2019 by SAE International in United States
Citation: Lanka, V., Heydinger, G., and Guenther, D., "Driver’s Response Prediction Using Naturalistic Data Set," SAE Int. J. Adv. & Curr. Prac. in Mobility 1(2):524-530, 2019, https://doi.org/10.4271/2019-01-0128.
Evaluating the safety of Autonomous Vehicles (AV) is a challenging problem, especially in traffic conditions involving dynamic interactions. A thorough evaluation of the vehicle’s decisions at all possible critical scenarios is necessary for estimating and validating its safety. However, predicting the response of the vehicle to dynamic traffic conditions can be the first step in the complex problem of understanding vehicle’s behavior. This predicted response of the vehicle can be used in validating vehicle’s safety.
In this paper, models based on Machine Learning were explored for predicting and classifying driver’s response. The Naturalistic Driving Study dataset (NDS), which is part of the Strategic Highway Research Program-2 (SHRP2) was used for training and validating these Machine Learning models. Various popular Machine Learning Algorithms were used for classifying and predicting driver’s response, such as Extremely Randomized Trees and Gaussian Mixture Model based Hidden Markov Model, which are widely used in multiple domains.
For classifying driver’s response, longitudinal acceleration vs lateral acceleration plot (Ax-Ay plot) was divided into nine different classes and selected Machine Learning models were trained for predicting the class of driver’s response. Performances of models for classification were tabulated and it is observed that Extremely Randomized Trees based model had better prediction accuracies in comparison with other models when fit using SHRP2 NDS data. The input features were reduced using dimension reduction techniques to reduce the computation time by over 70%.