Previous Abstract Return to Session C1 Next Abstract

Session C1: Sensor Aiding and Augmenting

Multiple Ultrasonic Aiding System for Car Navigation in GNSS Denied Environment
M. Moussa, A. Moussa, N. El-Sheimy, University of Calgary, Canada
Location: Windjammer

Nowadays, there are many attempts to enhance the navigation solution for Self-Driving Vehicles (SDVs) using low cost sensors with the overall objective of reducing the cost of the navigation sensors and hence reducing the overall price of SDVs. Global Navigation Satellite System/Inertial Navigation System (GNSS/INS) integration is the most common technologies for car navigation, however, the quality of the navigation states are usually deteriorated during GNSS signal outages because of the INS error large drift. Thus, aiding this integration with other low-cost sensors is necessary to bound such drift. Many previous researches used some sensors to aid the GNSS/INS integration such as odometer, Light Detection And Ranging (LIDAR), cameras, Radio Detection And Ranging (RADAR), etc. However, some of these sensors are very expensive in addition to their highly computational and processing requirements. Therefore, using low-cost sensors to aid the GNSS/INS integration becomes a very important research area to enhance the navigation of SDVs.
This paper introduces a novel approach for estimating the navigation states (position, velocity, and attitude) of ground vehicles in GNSS denied environment by integrating very low-cost ultrasonic sensors system with the Inertial Measurement Unit (IMU) using Extended Kalman Filter (EKF). This integration will limit the drift of low-cost Micro Electrical Mechanical Sensor (MEMS) during GNSS signal outage through a velocity update to provide an enhanced navigation estimates.
The ultrasonic sensor is usually used in ground vehicles as a ranging sensor for collision avoidance. However, here in this research, the ultrasonic is used as an aiding sensor for navigation. The Ultrasonic sensors are mounted on the body of the car facing both the rear right and rear left wheels to sense the range difference between the sensor and the spokes of the wheel to determine the angular velocity in revolution per seconds. i.e. the sensor used to measure the distance between the sensor and both the solid and the void parts of the rim. The output of the ultrasonic sensor will be pulse waves over time during the rotation of the wheel, where the minimum distance is determined when the sensor is facing the solid part of the rim, On the other hand, the maximum range is measured when the sensor is facing the voids of the wheel’s rim. The minimum and maximum ranges represent the bottom and the top of the pulse wave respectively.
However, the pulse width depends on the width of the solid and the void parts of the rim, the rotation rate of the wheel, and the sensor rate. While, the frequency (period) of the pulse depend on the number of the solid and void parts of the rim and the rotation rate of the wheel.
As a result, the time difference is calculated to determine the time taken between two successive signal spikes or drops to get the angular velocity of each wheel in revolution per second after dividing by the number of the solid parts of the rim. Finally, the speed is calculated as well as the distance travelled for each wheel, in which the speed can be used as a velocity update to aid the INS during GNSS signal outages.
Unfortunately, the ultrasonic raw data is contaminated with outliers and noise. Therefore, a minimum and maximum threshold are introduced to the ultrasonic data for both the left and right wheels to get rid of these outlier in addition to de-noising the signal. Real data was collected at the University of Calgary region using Pixhawk (Px4) which consists of Invensense MPU-6000 and a U-blox GPS which are mounted on the roof of Ford Focus car. In addition to, two ultrasonic sensors (HC-SR04) are connected to an embedded board ( Arduino Uno R4) in which these two sensors are mounted facing the left rear and right rear wheels within three to five centimeters distance from the solid part of the wheel’s rim. During the car motion, the ultrasonic sensors measure the ranges to both the solid and the voids part of the rim and a pulse wave is formed and then the angular velocity of each wheel is calculated through determining the period of the signal spikes or drops and dividing by the number of solid parts which are six in the case of Ford Focus wheel’s rim and then the forward speed is determined for each wheel.
Actual results show that aiding multiple ultrasonic sensor system to low cost MEMS provides more accurate navigation states than that provided by the MEMS navigation solution in a standalone mode. The position Root Mean Square Error (RMSE) is reduced from 22.35 meters to 2.43 meters during 30 seconds GNSS signal outage. While it is improved from 101.18 meters to 5.07 meters during 60 seconds GNSS outage. Moreover, the position states are enhanced from 345.24 meters to 14.45 meters during 120 seconds GNSS signal outage. While it improved from 584.47 meters to 22.27 meters during 180 seconds GNSS signal blockage.
In conclusion: the new proposed system is capable of enhancing the 3D RMS positioning errors for car navigation by around 89% during 30 seconds and around 95% for 60 seconds GNSS outage. Furthermore, the position navigation states are enhanced by more than 96 % during 120 and 180 seconds for INS in a standalone mode. This paper proves that the integration of the ultrasonic with INS has the advantage of being very low-cost sensor (less than 5 dollars), and runs on high data rate mode in addition to providing a redundant speed information as well as its simplicity for installation, data collection and data processing. Moreover, this system can be easily mounted on any car type to enhance its navigation during any GNSS blockage.



Previous Abstract Return to Session C1 Next Abstract