An Efficient and Novel Method of Evaluating the Impact of MEMS Inertial Sensor Errors on the Performance of Automotive Dead Reckoning

Tsung-Yu Chiou, An-Lin Tao, Pei-Shan Kao, Tzu-Hao Kuo, Li-Min Lin, and Pei-Yu Huang

Abstract: The purpose of this work is to assess the impact of MEMS inertial sensor errors on the performance of Automotive Dead Reckoning (ADR). The inertial sensor bias includes deterministic and stochastic errors. In this work, we focus on the ADR performance impact by a deterministic error due to temperature change and the stochastic portion of the sensor bias errors. Traditionally, people use a simple model stating that a gyro bias introduces a cubic error in the position estimate. However, the modeled positioning error can be too pessimistic, i.e., the modeled positioning error would be larger than the error of the positioning outputs from the GNSS/INS system. For the automotive dead reckoning, there are vehicle related constraints which are helpful for sensor bias estimation and calibration in run. In this case, the dead reckoning error is much smaller than the error modeled by the aforementioned simple model. We have built a test bed which allows us to determine the dead reckoning performance given the sensor bias and the extra bias due to temperature change, and our GNSS/INS integration solution. We will present the results in terms of the error of distance traveled (EDT) to address the quality of 4 off-the-shelve MEMS inertial sensors. The results will describe the EDT performance given the GNSS/INS solution with the sensor bias at room temperature and the sensor bias having temperature change from -30°C to 70°C. Finally, the field trial performance will be presented in various scenarios such as driving through a 13 km tunnel, indoors, and in urban areas. The average EDT of 2% can be achieved with the room temperature condition and the average EDT of 5% can be achieved with a temperature change rate of 1°C/second, from -30°C to 70°C and vice versa. When designing a GNSS/INS solution for an ADR, a key step is to determine the qualified inertial sensor which has the minimum cost. Checking the parameters in the specification sheet is usually the first step but the relationship between the sensor specifications to a designed GNSS/INS solution is not straightforward. The overall ADR performance depends on both the sensor quality and the algorithm of GNSS/INS solution. We like to know, given the designed algorithm, what are the qualified sensors and the corresponding ADR positioning performance. We can do the field trial with the various sensors directly, but the cost of doing the experiments is high. Therefore, we propose a method to do the ADR performance evaluation based on different sensor bias conditions. The basic idea of the method is to use the designed GNSS/INS software to process the real static MEMS inertial sensor raw data. The output performance of the ADR would represent the total effects of the algorithm and the sensor errors. Processing the static MEMS inertial sensor data is to mimic the scenario of driving in a straight tunnel with a constant vehicle speed. Since the algorithm applies non-holonomic constraints (NHC) and the vehicle speed aiding when driving in no GNSS signal condition, the sensor bias will be estimated by the Kalman filter through the measurement update from the NHC and the vehicle speed. The overall effects are nonlinear. Thus, we use the aforementioned method to evaluate the final impact from the sensor errors. Furthermore, the extra sensor bias due to the temperature change is added into the recorded static MEMS inertial sensor data to mimic the driving in a constant speed with the temperature change. The extra sensor bias profile of a specific sensor is obtained by doing the thermal calibration in a thermal chamber. The profile is in terms of the extra sensor bias versus sensor temperature from -30°C to 70°C. Having this extra sensor bias profile, we can generate a sensor bias rate profile given a temperature change rate. For example, a car is parked under the open sky on a sunny day for hours. The interior temperature of the car and the sensor temperature could be as high as 50°C or higher. When the driver turns on the power and the air condition, the temperature could change from 50°C to 25°C in minutes. On the other hand, a car on a winter day would have temperature changing from -30°C to 25°C in minutes. To have a stringent condition, we use the temperature change rate of 1°C/second to generate the temperature change profile. Given the intrinsic sensor bias plus the extra bias due to the temperature change, a constant vehicle speed signal, and the NHC in the GNSS/INS software, we can evaluate the final ADR positioning performance. A first order Gauss-Markov model is implemented in the GNSS/INS software. The parameters of the Gauss-Markov are obtained by doing the Allan variance analysis. By doing the above actions, it is relatively having less efforts when qualifying a new MEMS inertial sensor for a designed ADR solution. What we need to do are collecting several hours of the static sensor raw data, performing the thermal chamber experiments, and doing the Allan variance analysis. In summary, this work presents a process of qualifying an inertial sensor given a designed ADR solution. The results show the minimum required sensor specifications and the quality ranking of 4 off-the-shelve MEMS inertial sensors. The ADR performance is optimized according to the sensor error characteristics. The EDT performance can be within 2% in average for indoor and tunnel cases and the EDT can be in 5% with high temperature change rate. The ADR solution has been built on an embedded system and running in real time. After qualifying the MEMS inertial sensor and characterizing the sensor errors, the ADR solution is designed to provide very good performance in various challenging environments, such as a large basement, a high parking tower, a long tunnel, and urban areas. The best performance, for example, can be achieved to an positioning error of 40 m after driving through a 13 km long tunnel (i.e., EDT=0.31%). The best positioning performance of driving in a basement for 20 min is 7 m error at the exit of the basement (i.e., EDT=0.19%).
Published in: Proceedings of the 31st International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2018)
September 24 - 28, 2018
Hyatt Regency Miami
Miami, Florida
Pages: 2051 - 2073
Cite this article: Chiou, Tsung-Yu, Tao, An-Lin, Kao, Pei-Shan, Kuo, Tzu-Hao, Lin, Li-Min, Huang, Pei-Yu, "An Efficient and Novel Method of Evaluating the Impact of MEMS Inertial Sensor Errors on the Performance of Automotive Dead Reckoning," Proceedings of the 31st International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2018), Miami, Florida, September 2018, pp. 2051-2073.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In