Driver’s response prediction using large Naturalistic Data Set

Paper #:
  • 2018-01-0507

  • 2018-04-03
Evaluating the safety of Autonomous Vehicles (AV) is a challenging problem, especially when it is having dynamic interactions with other vehicles. 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 is the first step in the complex problem of safety evaluation. In this paper, methods based on Machine Learning are explored for predicting and classifying drivers' response. Naturalistic Driving Study dataset (NDS), which is part of Strategic Highway Research Program-2 is used for training and validating these Machine Learning models. Two popular Machine Learning Algorithms are used for predicting driver's intent & response, namely 1) Extremely Randomized Trees and 2) Gaussian Mixture Model based Hidden Markov Model, which are widely used in multiple domains. For classifying driver's intent, driver's response at a scenario is categorized into 3 categories: 1. Lane Keeping, 2. Lane changing 3. Emergency Lane changing. Also, driver's response, which is lateral and longitudinal acceleration is predicted. For fitting driver's response to the Machine Learning models, the NDS is split into testing, training and validating data. Model parameters are identified using testing & training data and are validated against validation data set. Detailed analysis of the results is done for estimating performance of both algorithms. Comparison against traditional modeling approaches is done and results show these algorithms are better at predicting driver's response than traditional model-based approaches.
SAE MOBILUS Subscriber? You may already have access.
Attention: This item is not yet published. Pre-Order to be notified, via email, when it becomes available.
Members save up to 36% off list price.
HTML for Linking to Page
Page URL

Related Items

Technical Paper / Journal Article
Training / Education