Lane Keeping Assistance with Learning-Based Driver Model and Model Predictive Control

Stéphanie Lefèvre (1, 2) , Yiqi Gao (1) , Dizan Vasquez (2) , H. Eric Tseng (3) , Ruzena Bajcsy (1) , Francesco Borrelli (1)

1 EECS - Department of Electrical Engineering and Computer Science [Berkeley]
2 E-MOTION - Geometry and Probability for Motion and Action
3 Ford Research Laboratories

This paper proposes a novel active Lane Keeping Assistance Systems (LKAS) which relies on a learning-based driver model. The driver model detects unintentional lane departures earlier than existing LKAS, and as a result the correction needed to keep the vehicle in the lane is smaller. When the controller has control of the car, the driver model estimates what the driver would do to keep the car in the lane, and the controller tries to reproduce that behavior as much as possible so that the controlled motion feels comfortable for the driver. The driver model combines a Hidden Markov Model and Gaussian Mixture Regression. The controller is a Nonlinear Model Predictive Controller. The results obtained with real data show that our driver model can reliably predict lane departures. The controller is able to keep the car in the lane when there is a risk of lane departure, and does so less intrusively than existing LKAS. Topics / Active safety and driver assistance systems, Driver modeling

Domaines: Robotique [cs.RO]

Stéphanie Lefèvre, Yiqi Gao, Dizan Vasquez, H. Eric Tseng, Ruzena Bajcsy, et al.. Lane Keeping Assistance with Learning-Based Driver Model and Model Predictive Control. 12th International Symposium on Advanced Vehicle Control, 2014, Tokyo, Japan. ⟨hal-01104458⟩