Road Map Generation and Vehicle Localization for Automated Driving

Abstract

This dissertation aims to present precise and cost-efficient mapping and localization algorithms for autonomous vehicles. Mapping and localization are ones of the key components in autonomous vehicles. The major concern for mapping and localization research is maximizing the accuracy and precision of the systems while minimizing the cost. For this goal, this dissertation proposes a road map generation system to create a precise and efficient lane-level road map, and a localization system based on the proposed road map and affordable sensors.

In chapter 2, the road map generation system is presented. The road map generation system integrates a 3D LIDAR data and high-precision vehicle positioning system to acquire accurate road geometry data. Acquired road geometry data is represented as sets of piecewise polynomial curves in order to increase the storage efficiency and the usability. From extensive experiments using a real urban and highway road data, it is verified that the proposed road map generation system generates a road map that is accurate and more efficient than previous road maps in terms of the storage efficiency and usability.

In chapter 3, the localization system is presented. The localization system targets an environment that the localization is difficult due to the lack of feature information for localization. The proposed system integrates the lane-level road map presented in chapter 2, and various low-cost sensors for accurate and cost-effective vehicle localization. A measurement ambiguity problem due to the use of low-cost sensors and poor feature information was presented, and a probabilistic measurement association-based particle filter is proposed to resolve the measurement ambiguity problem. Experimental results using a real highway road data is presented to verify the accuracy and reliability of the localization system.

In chapter 4, an application of the accurate vehicle localization system is presented. It is demonstrated that sharing of accurate position information among vehicles can improve the traffic flow and suppress the traffic jam effectively. The effect of the position information sharing is evaluated based on numerical experiments. For this, a traffic model is proposed by extending conventional SOV traffic model. The numerical experiments show that the traffic flow is increased based on accurate vehicle localization and information sharing among vehicles.