Identification of intoxicated driving using steering wheel signals and improving lateral control in semi-autonomous and autonomous vehicles

The studies reported in this thesis focus on specific problems in detection of intoxicated driving, improving the performance of the vehicle when an intoxicated driver is controlling the vehicle, and designing autonomous lateral controllers. In the first phase of this study, we apply system identification techniques on the steering wheel control behavior of the driver to present two models to describe the behaviors of sober and drunk drivers. Then we use these models and online identification methods to detect intoxicated driving from steering wheel data and vehicle lateral position. In the second part of this thesis, we present the idea of improving the steering action of intoxicated drivers by adding serial and parallel controllers to the system while the driver is in the loop. In the first proposed algorithm, the steering signal coming from the steering wheel is fed to a serial controller. The output of the controller becomes the actual steering of the car. In the second suggested algorithm, the output of an independent lateral controller is added to the control signal generated by the human driver. In the third phase, several look-ahead lateral controllers are designed to maintain the vehicle in the center of the lane when the driver is removed from the system. Among the designed controllers are a novel, simple fused neural-network controller, introduced by our group, and a recently introduced robust adaptive controller which applies L_1 adaptive control theory on vehicles for the first time. The designed controllers are tested in challenging scenarios including wind gusts, road banking, icy roads, vehicle parameter uncertainties, and measurement noise, all present at the same time. Finally, longitudinal controllers are studied, designed, and combined with the previously designed lateral controllers to complete the control subsystem of autonomous vehicles.