Estimating Vehicle State by GPS/IMU Fusion with Vehicle Dynamics

Abstract

In this thesis, an implementation of a low-cost inertial navigation system with high integrity and reliability is proposed. The proposed system can be utilized for warning drivers in land vehicle applications of an approaching dangerous situation. The solution uses Inertial Measurement Unit (IMU) to provide the vehicle's state by position, velocity and attitude readings. However, these readings quickly diverge from the correct state of the vehicle in a fashion that makes them unreliable to use. To solve this issue, a GPS is fused with the Inertial Navigation System (INS). This fusion reduces the estimation errors considerably. Two approaches are proposed for sensor fusion. In the first approach, the extended Kalman filter (EKF) is used to optimally fuse the sensor readings. In order to reduce the computational complexity of the EKF algorithm, the extended information filter (EIF) is used as an alternative state estimator. The performance of these two approaches is analyzed and compared in this thesis. Given that the proposed system is intended for use in land vehicle applications, nonholonomic velocity constraints and encoder velocity readings are fused with the algorithm to enhance the accuracy of the estimates. The filter uses a fifteen-element state. This state is composed of the errors in position, velocity, quaternions, accelerometers' biases and gyroscopes' biases. To enhance the integrity of the estimated state, the Limited Memory Noise Estimation (LMNE) algorithm is developed and proposed to detect possible bias in the encoder measurement. Extensive experimental tests were conducted to verify the accuracy of the proposed algorithms. The obtained results were compared with a commercial off-the shelf (COTS) MIDG solution. The results obtained from the EKF and the EIF were comparable in accuracy, but EIF needed less processing time compared to the EKF. The estimation accuracy was improved when the nonholonomic velocity constraints were used. The experimental results also showed that the LMNE was effective in detecting and identifying biases that were intentionally added on the encoder's velocity readings. Search Terms: Kalman filter, information filter, velocity constraints, nonholonomic constraints, fault detection, noise estimation.