Exploring Advanced Control Mechanisms for an Autonomous Scale-Model Vehicle

Applonie, Robert Reed
Journal Title
Journal ISSN
Volume Title

Control mechanisms for autonomous vehicles have been studied for decades, but recently they attract more research due to the boom of self-driving vehicles. This thesis explores the design and implementation of the key control strategies of autonomous vehicles.

A linear PID speed control was developed with an emphasis on model-based design. A dy-namometer-like device was constructed to establish a transfer function between motor speed command (input) and actual vehicle speed (output). The PID controller was then integrated with the vehicle transfer function to determine appropriate gain factors for optimum performance. The controller was applied to the autonomous scale-model vehicle and performance of the system was recorded and verified against the actual tracking error on a real track. The PID controller provided a baseline to compare against other advanced control strategies.

A nonlinear steering control law was developed that encompassed the lateral and orientation errors based on an on-board camera at a look-ahead distance which, along with the controller gain, was tuned dynamically as a function of velocity of the vehicle. Stability of the control law was demonstrated by Lyapunov analysis and the relationship of look-ahead distance and control gain were also determined. A simulation framework was developed to demonstrate the effectiveness of the nonlinear controller, determine the effects of camera and actuator delay on the system, and to find the relationship between speed and look-ahead distance that minimizes tracking error.

Finally, a proof-of-concept for a deep-learning approach for steering control was tested. The deep-learning study examined a classification model with the goal of assisting the computer vision algorithms; it also developed a regression model that calculated the requisite steering angle directly. The classification model, while interesting, was abandoned in favor of the regression model, which predicted steering angle values with a lower error rate than actual remote control human driving.

This item is available only to currently enrolled UTSA students, faculty or staff.
Autonomous vehicles, Lyapunov methods
Electrical and Computer Engineering