Staying Inside the Lines: Vehicle Agnostic Path Following Using Cascaded Adaptive Control
Type of DegreeMaster's Thesis
MetadataShow full item record
This thesis presents a vehicle agnostic steering controller for path following. Many active safety systems, such as collision avoidance and lane centering, as well as all SAE Level 2+ autonomous vehicles, rely on a lateral controller to follow a desired path. The vehicle agnostic path following controller presented in this thesis is comprised of two adaptive controllers in a cascaded architecture, with the outer loop controlling the path dynamics and the inner loop controlling the vehicle dynamics. First, some commonly used tire models and lateral vehicle models are introduced. Next, a sensitivity analysis is performed on a variety of lateral controllers under the influence of model uncertainties. This analysis is used to develop the vehicle agnostic path following controller, which is tested in simulation at multiple velocities on a sedan, an SUV, a pickup truck, and a minivan. In simulation, the controller is able to adapt to each platform and achieve good lane keeping performance around a curvy track at a wide range of speeds. The controller is then implemented and validated in real-time on a Lincoln MKZ, a Class-8 Peterbilt 579 cab, and a Peterbilt 579 with a loaded trailer. During a double lane change maneuver, the vehicle agnostic path following controller maintains maximum path tracking errors of approximately half a meter with all three of these setups. The controller is also implemented on a 1/10th scale RC car using a vision-based lane centering system to generate the reference path. Even on this scaled platform, the controller is able to follow the lane lines at multiple speeds, with maximum lookahead errors of 15 cm. Overall, the controller is shown to perform well on four simulated platforms and four experimental platforms at a range of longitudinal speeds, demonstrating the flexibility of the path following controller.