Motion Planning using Positively Invariant Sets on a Small-Scale Autonomous Vehicle

Abstract: Self-driving technology has the opportunity to increase safety in automotive transportation by reducing the impact of human error. Motion planning is a key component of an autonomous system, responsible for providing reference trajectories and paths that the vehicle should follow. This thesis studies a motion-planning algorithm based on positively invariant sets. The focus is on design, implementation, and evaluation of the algorithm on a small-scale ground-vehicle robot platform. By incorporating a gain-scheduling approach into the motion planner, guaranteed safe reference trajectories, capable of navigating the vehicle in a dynamic environment of static and moving obstacles, can be computed for time-varying velocities. This thesis also deals with sensor-fusion aspects for autonomous vehicles. Through a localization system based on an Extended Kalman Filter (EKF), reliable and robust state estimates can be obtained from inertial sensor data, without the use of an external positioning system. It is shown that the motion-planning algorithm together with the localization system is capable of performing safe overtaking maneuvers for time-varying velocities. Simpler urban driving scenarios involving traffic signs and intersections are used to illustrate the ability of the proposed motion-planning algorithm to also handle more complex driving scenarios. By using laser scans from Light Detection and Ranging (LIDAR) equipment, it is shown that obstacles can be detected and avoided in real-life driving experiments.