Abstract
Key Words Out-of-sequence measurement, Bayesian filter, Localization, GPS, GPS time pulse
Research and Engineering Questions/Objectives
For autonomous vehicles, Bayesian filter-based localization has been researched in order to estimate vehicle positions. The GPS is commonly used as a measurement of the Bayesian filter. However, the GPS has numerous errors, including ionosphere, troposphere, and ephemeris errors. To correct such GPS errors, the differential GPS has been used, but, the correction process is time-consuming. For this reason, measurements are delayed, and these delayed measurements are called an Out-Of-Sequence Measurement (OOSM). Consequently, the OOSM degrades the localization performance. In this paper, a fast OOSM processing algorithm is introduced to improve localization performance and to stabilize vehicle control.
Methodology
The OOSM processing algorithm utilizes additional information to eliminate the negative effects of the OOSM. This additional information reflects the measurement arrival time, tk, which does not contain the delay. In the case of GPS, the “GPS time pulse” plays a role as a part of this additional information. The GPS time pulse is counted whenever GPS signals arrive at the GPS receiver. Using the counted pulse, the time delay, td, is calculated precisely. When GPS measurements are fused with other information at the time of tk+td due to time delay, a “retrodiction” is performed from tk+td to tk, based on the Bayesian filter. During the retrodiction, the position of the ego-vehicle is stochastically estimated in reverse. This “reverse estimation” provides the estimated position at the time tk. After the retrodiction, delayed measurements are utilized to update the reversely-estimated position because the measurements are generated at the time tk.
Results
In this study, the GPS time pulse is used as timing information to remove the negative effects of the OOSM. With the algorithm presented in this paper, the trajectory of the estimated position is smoother than the trajectory without the GPS time pulse. The smooth trajectory of the estimated position means that localization performance has been enhanced. For autonomous vehicles, the reliable localization system stabilizes vehicle control. In particular, in the case of high dynamics conditions (e.g., 100kph), the benefit of the presented processing algorithm is remarkable.
Limitations of this study
The processing algorithm presented in this paper is applied only when the measurement arrival time, tk, is known by the time-indicating information. Without this additional information, the retrodiction is less accurate and time-consuming.
What does the paper offer that is new in the field including in comparison to other work by the authors?
In previous researches, time delay must be assumed because the arrival time of the measurements is unknown. When the arrival time is precisely known from timing information, computational efforts can be reduced. As a result, the processing procedure is faster than previous OOSM processing algorithms.
Conclusions
As an information fusion technology, the OOSM should be considered because time delay occurs unavoidably. If the exact measurement arrival time is known, The OOSM is processed easily and quickly. In this paper, the GPS time pulse is utilized as the timing information for GPS measurements.