The purpose of this research is to construct a self-driving bicycle that can balance itself and automatically track a designated trajectory in campus environments. For balancing, a lower-level controller is designed based on the dynamic model of the bicycle. It allows the bicycle to achieve lateral stability and cornering action with robustness to speed variations. The design methodology adopts a linear-parameter-varying (LPV) approach by firstly decomposing the dynamic model into a convex combination of four linear subsystems with time-varying coefficients and then solving a set of linear matrix inequalities (LMI’s) to compute the gain matrix for robust state feedback. For trajectory tracking, a high-level controller is design using the similar LPV approach. It allows the bicycle to robustly follow a pre-generated virtual vehicle motion on a given path regardless of the speed and yaw rate changes of the virtual vehicle along the path. To provide feedback signals for both controllers, sensor fusion algorithms are proposed to fuse GPS, wheel speed and IMU signals to obtain bicycle posture and localization information. The control system is verified both numerically and experimentally on a prototype bicycle. Particularly, the experiment shows that the self-driving bicycle can follow the testing route in campus with RMS errors less than 18cm.
