AbstractsEngineering

Accurate localization is essential to the safe and effective functioning of an autonomous vehicle. In an autonomous valet parking system, the vehicle must be able to estimate its position in the global coordinate frame in order to plan its path and avoid obstacles. Furthermore, precise localization information is necessary for feedback for the control algorithms governing both general parking lot navigation as well as various parking maneuvers.This thesis explores the application of a real-time LIDAR-based landmark sensing scheme combined with a popular simultaneous localization and mapping method known as FastSLAM. The sensing algorithm extracts vertical objects from a 3D Velodyne lidar scan by applying a connected components algorithm to a 2D occupancy grid that is built from the scan. These landmarks are associated robustly from frame to frame in FastSLAM, which is essentially a Rao-Blackwellized particle filter where each particle uses 2D Kalman Filters to estimate the positions of known landmarks.The localization algorithm is tested using data collected from driving and performing parking maneuvers in a typical parking lot. Simulated data is also generated to verify the algorithm and to test its ability to handle varying levels of sensor error and landmark density. Advisors/Committee Members: Ozguner, Umit (Advisor).