Multiple Faulty GNSS Measurement Exclusion Based on Consistency Check in Urban Canyons

Abstract

Sensors play important roles for autonomous driving. Localization is definitely a key one. Undoubtedly, global positioning system (GPS) sensor will provide absolute localization for almost all the future land vehicles. In terms of driverless car, 1.5 m of positioning accuracy is the minimum requirement, since the vehicle has to keep in a driving lane that usually wider than 3 m. However, the skyscrapers in highly-urbanized cities, such as Tokyo and Hong Kong, dramatically deteriorate GPS localization performance, leading more than 50 m of error. GPS signals are reflected at modern glassy buildings, which caused the notorious multipath effect. Fortunately, the number of navigation satellite is rapidly increasing in a global scale, since the rise of multi-global navigation satellite system. It provides an excellent opportunity for positioning algorithm developer of GPS sensor. More satellites in the sky imply more measurements to be received. Novelty, this paper proposes to take advantage of the fact that clean measurements (refers to line-of-sight measurement) are consistent and multipath measurements are inconsistent. Based on this consistency check, the faulty measurements can be detected and excluded to obtain better localization accuracy. Experimental results indicate that the proposed method can achieve less than 1-m lateral positioning error in middle urban canyons.