Accurate Localization and Motion Planning for Autonomous Driving at Urban Intersections

Prarthana Bhattacharyya, Yanlei Gu, Jiali Bao, and Shunsuke Kamijo

Abstract: Autonomous driving technologies have necessitated the need for effective navigation systems to drive through complex urban intersections. The key to efficient motion planning and navigation is accurate self-localization. In addition to this, detecting, tracking, lane-localizing and estimating the future intent of surrounding traffic participants with respect to the ego-vehicle as well as the road-lane structure is also necessary to establish vehicle cooperation for navigation. This paper proposes a method to accurately self-localize an ego-vehicle at an urban intersection. In addition, the paper also proposes to localize and understand the behavior of the surrounding traffic participants in order to help the ego-vehicle navigate through complex urban intersections. Ego-localization has been dominated mainly by global navigation satellite systems (GNSS) sensors. However in urban areas, GNSS performance is severely degraded due to Non-Line-Of-Sight (NLOS) and multipath effects. In order to rectify the pseudo-range error caused by NLOS and multipath, our research group proposed to improve 3D GNSS methods by taking advantage of 3D building maps. However our previous positioning system still suffers from error along the longitudinal direction and the heading direction, which is significant for localizing ego-vehicle and estimating the position of other traffic participants. Also since lane-markings are not visible at intersections, vision based lane detection will fail in these cases and will lead to large positioning error. In order to solve these problems, this paper proposes to integrate 3D building matching with our previous 3D GNSS/INS/lane detection method in order to improve the longitudinal position and orientation estimation of the ego-vehicle. Furthermore, accurate localization is extremely important to detect the behavior of surrounding traffic participants. Thus, the detected and filtered state estimates of the surrounding vehicles are formulated into a Dynamic Bayesian Network (DBN) in order to infer their behavior at the intersection. This situation recognition, which is the key to motion planning, is addressed in our paper using DBN. The construction of the DBN is based on learning from the data collected from the real traffic scenes. The contribution of this paper is to achieve an accurate self-localization result for ego- vehicle in urban intersection while simultaneously being able to localize and estimate the behavior of surrounding traffic participants.
Published in: Proceedings of the ION 2017 Pacific PNT Meeting
May 1 - 4, 2017
Marriott Waikiki Beach Resort & Spa
Honolulu, Hawaii
Pages: 447 - 458
Cite this article: Bhattacharyya, Prarthana, Gu, Yanlei, Bao, Jiali, Kamijo, Shunsuke, "Accurate Localization and Motion Planning for Autonomous Driving at Urban Intersections," Proceedings of the ION 2017 Pacific PNT Meeting, Honolulu, Hawaii, May 2017, pp. 447-458. https://doi.org/10.33012/2017.15026
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In