Development of autonomous electric-vehicle using resilience logic based localization and obstacle avoidance in complex urban environment

Abstract

In the recent years, there have been a significant amount of driver-less transportation projects to fulfill autonomous vehicles on urban or dedicated environment, such as terminal, airports and an amusement park. However, there are still innumerable challenges when autonomously driving in a complex urban environment such as downtown and campus scenarios. Unfortunately, localization based on Global Navigation Satellite System(GNSS) and dead reckoning(DR) using GPS and IMU sensors do not always guarantee in the such complex urban environment. Thus, a localization system was implemented to reduce the positioning error by correcting the lateral position error of the vehicle in the area where lane detection is feasible. However, there were still remained limitations in that both GNSS/DR and lane detection were not possible. As a result, an alternative approach was necessary such as 3d pointcloud map based localization to compensate the position error. On the other hand, to come up with 3d map, it also has a limitation in a large-scale map. Therefore, in this study, a resilience logic based localization system was proposed in order to consider a redundancy in the navigation system and to take a advantage of strength of each navigation system. Furthermore, to enable to drive in the complex urban area, it was integrated that obstacle avoidance system using a number of path candidates which were generated based on the road model. Therefore, it was implemented changing lanes and choosing the collision free path candidates. In addition, in order to verify the proposed system, autonomous electric-vehicle, of which capacity approximately is 12-14 passengers, is developed and a number of experiments were conducted in the complex urban environment. In this paper, we analyzed the result and limitation of each navigation system and represented necessity of resilience logic based localization system and road model based obstacle avoidance. In addition, we generated 3d pointcloud map in the KAIST campus, and the coordinates of each system was unified between real-world and a generated 3d pointcloud map in order to integrate the navigation systems. Finally, the demonstration was carried out without any human intervention at the events in the KAIST and it was shown that the developed system was successfully performed autonomous driving the complex urban area.