Development of urban self-driving vehicle using lane detection on the road and localization in intersection with lane map

Abstract

Recently, there have been great advances in autonomous driving technology, which would be ready for everyday use in near future. In order for self-driving cars to become viable, it must be fully ready to perform safely and reliably. Applying the technologies based on simulation directly, it is difficult to make vehicles that can run itself in actual road environment. The biggest issue is localization. The precise location can be obtained by using high performance GNSS (DGPS), but there are also limitations in obtaining solutions in signal block situations or multipath errors due to buildings and tunnels in urban environments. In order to solve this problem, there is a method of correcting the position information by integrating with the IMU, but cumulative error still occurs. In addition, it can improve position accuracy by correcting the position through map matching method. However, there is a disadvantage that large resources are required for data storage and processing in a large-scale environment. Furthermore the buildings used for the map matching are varied in the urban environment and a few in the suburbs. In this thesis, autonomous driving mode are divided into three parts in order to develop autonomous vehicles that can run in an actual environment. Each mode are road following, localization at an intersection and obstacle avoidance. Road following mode enables autonomous driving even when the location information is inaccurate through lane recognition on the road. In addition to Lane keeping assistance system (LKAS) and adaptive cruise control (ACC) functions that can be used on roads without bifurcation, such as highways, Road following also implements the ability to travel in narrow alleys. Localization at an intersection is a mode used to obtain accurate location information at an intersection. Localization is performed by using only a map of a specific area, which is an intersection, without using a map in all road areas. This has the advantage of saving resources in data storage and processing in a largescale environment. Landmark required for map matching can be used robustly even in a building-less environment with the road mark instead of buildings. Finally, obstacle avoidance is a mode that operates to safely avoid obstacles at low speeds in a situation where the vehicle cannot proceed without avoidance maneuvers. This thesis deals with the development of autonomous vehicles in real road environment through the implementation of above three modes.