Yukihiko Ono

GNSS fault detection with unmodeled error

Graphical Abstract Abstract This paper provides a method for ‘fault detection’ on a GNSS/INS positioning system with unmodeled errors. Many GNSS/INS positioning systems apply testing with Maharanobis distance for fault detection of GNSS position. However, this method requires an accurate system model. If there are some inaccuracies in the model, we struggle to adjust the best parameters on the system model by trial and error. This fact sometimes reduces extendability of the system. Thus, we advance an unmodeled error discerning method focused on non-stationarity in residuals between observation time-series and estimated values. Our proposed method provides a stable period for observation time series. In that period, we can correctly calculate the covariance of unmodeled error and also the predisposition of unmodeled error to displace the origin of Mahalanobis distance to an appropriate point. This new procedure gives a better distribution for testing for fault detection than the conventional procedure. We verify our proposed fault detection in real-world measurement with a test vehicle on an irregular ground test circuit. The results confirmed that our proposed method improved the accuracy of localization. The proposed method will help us to apply positioning systems easily to other vehicles.

Self-localization with ultrasonic sensor array

We have developed a positioning system that uses a phased array sensor to measure the distance and direction of multiple landmarks relative to the position of a mobile robot so that the robot can self-localize. The positioning system can control the strength and the width of the beam from the phased array sensor and observe the multiple landmarks at the same time. By using the distance and direction of plural landmarks, our positioning system is able to use various self-localizing methods. A main part of our work has been incorporating our various techniques for making a robot positioning error smaller into a system in which the most suitable self-localizing method is selected on the basis of sensitivity, which is defined as the ratio between localizing and sensor errors. We propose using a set of indices to evaluate the accuracy of self-localizing methods. The indices are derived from the sensitivity. We used these indices to compare the accuracy of three methods for the self-localizing of a mobile robot using landmarks, and we demonstrated a rational way to minimize a localizing error.