Enhanced Full-state Estimation and Dynamic-model-based Prediction for Road-vehicles

In this paper, we address the enhanced state estimation and prediction system for automobile applications by fusing relatively low-cost and noisy Inertial Navigation System (INS) sensing with Global Positioning System (GPS) measurements. An unscented Kalman filter is used to merge multi-rate measurements from GPS and INS sensors together with a highfidelity vehicle-dynamics model for state-predictions. The highfidelity motion model (including suspension-effects) for the vehicle motion trajectory on uneven terrain is captured by a 20state system of nonlinear differential equations. Computer simulation results illustrate the effectiveness of sensor-fusion (building upon the merger of an inexpensive INS sensing with GPS based measurements) to accurately estimate the full system-state. The relative ease of implementation, accuracy and predictive performance with low-cost sensing will facilitate its use in various electronic control and safety-systems, such as Electronic Stability Program, Anti-lock Brake Systems, and the Lateral Dynamic Stability Control. INTRODUCTION Modern-day automobiles have seen tremendous improvements in reliability and safety due to incorporation of intelligent systems such as Anti-lock Braking System (ABS), Electronic Stability Program (ESP) and Lateral Dynamic Stability Control (DSC). Most such vehicle-control systems integrate noisy and biased sensor measurements or utilize model-based estimators in order to obtain accurate and up-todate critical vehicle state information such as sideslip [1, 2, 3], longitudinal and lateral velocities or roll angle [4, 5]. The nonavailability of these crucial vehicle states (at adequate rates) often poses a significant challenge for implementation of appropriate advanced vehicle-control algorithms. Similarly, steer-by-wire systems require the entire complement of current states for full state-feedback [6]. However, in current vehicles not all the necessary states can be accurately measured which necessitates implementation of the appropriate filtering and estimation methodologies. A popular technique in estimating the vehicle states is the direct use of the data provided by INS or GPS measurements. However, such data tends to be noisy due to the measurement uncertainty which, if ignored, can result in erroneous estimation. In case of INS sensing, unwanted measurements of the road grade and bank angle further degrades the quality of the sensed data [7, 8]. An alternate approach is using physical vehicle model as an observer sensitive to changes in the vehicle parameters [9, 10]. However, prior to a detailed discussion, it is noteworthy that most of the literature uses the relativelysimplistic linear-bicycle model with linear-tire models owing to ease of implementation and fewer parameters to specify. In several works, for example [1, 11, 12, 15, 16], the measurements provided by GPS are fused with the INS sensing. This enable to use low-cost and high rate INS-based measurements while noise effects are compensated by fusing low-rate (but drift-free) GPS data. In fact, when GPS signal is Aliakbar Alamdari Mechanical and Aerospace Engineering, SUNY at Buffalo Buffalo, NY, 14260 [email protected] Javad Sovizi Mechanical and Aerospace Engineering, SUNY at Buffalo Buffalo, NY, 14260 [email protected] Venkat N. Krovi Mechanical and Aerospace Engineering, SUNY at Buffalo Buffalo, NY, 14260 [email protected]