Date of Award
Doctor of Philosophy in Electrical Engineering - (Ph.D.)
Electrical and Computer Engineering
Richard A. Haddad
Janice Rhoda Daniel
Advanced driver assistance systems (ADAS) have been developed to help drivers maintain stability, improve road safety, and avoid potential collision. The data acquisition equipment that can be used to measure the state and parameter information of the vehicle may not be available for a standard passenger car due to economical and technical limitations. This work focuses on developing three technologies (longitudinal tire force estimation, driver behavior classification and lateral control) using low-cost sensors that can be utilized in ADAS.
For the longitudinal tire force estimation, a low cost 1Hz positioning global system (GPS) and a steering angle sensor are used as the vehicle data acquisition equipment. A nonlinear extended two-wheel vehicle dynamic model is employed. The sideslip angle and the yaw rate are estimated by discrete Kalman Filter. A time independent piecewise optimization scheme is proposed to provide time-continuous estimates of longitude tire force, which can be transferred to the throttle/brake pedal position. The proposed method can be validated by the estimation results.
Driver behavior classification systems can detect unsafe driver behavior and avoid potentially dangerous situations. To realize this strategy, a machine learning classification method, Gaussian Mixture model (GMM), is applied to classify driver behavior. In this application, a low cost 1Hz GPS receiver is considered as the vehicle data acquisition equipment instead of other more costly sensors (such as steering angle sensor, throttle/brake position sensor, and etc.). Since the driving information is limited, the nonlinear extended two-wheel vehicle dynamic model is adopted to reconstruct the driver behavior. Firstly, the sideslip angle and the yaw rate are calculated since they are not available from the GPS measurements. Secondly, a piecewise optimization scheme is proposed to reproduce the steering angle and the longitudinal force. Finally, a GMM classifier is trained to identify abnormal driver behavior. The simulation results demonstrated that the proposed scenario can detect the unsafe driver behavior effectively.
The lateral control system developed in this study is a look-down reference system which uses a magnetic sensor at the front bumper to measure the front lateral displacement and a GPS to measure the vehicle's heading orientation. Firstly, the steering angles can be estimated by using the data provided by the front magnetic sensor and GPS. The estimation algorithm is an observer for a new extended single-track model, in which the steering angle and its derivative are viewed as two state variables. Secondly, the road curvature is determined based on the linear relationship with respect to the steering angle. Thirdly, an accurate and real-time estimation of the vehicle's lateral displacements can be accomplished according to a state observer. Finally, the closed loop controller is used as a compensator for automated steering. The proposed estimation and control algorithms are validated by simulation results. The results showed that this lateral steering control system achieved a good and robust performance for vehicles following or tracking a reference path.
Yang, Jing, "Driver behavior classification and lateral control for automobile safety systems" (2012). Dissertations. 303.