Interacting Multiple Model Kalman Filtering for Optimal Vehicle State Estimation

Abstract

In this paper, we propose a new method for optimal vehicle condition estimation by utilizing low-cost sensor fusion using in-vehicle sensors and a standalone Global Positioning System (GPS). Estimation targets are non-measurable vehicle conditions such as vehicle side slip angles and tire cornering stiffness values of front and rear axles. Interacting Multiple Model (IMM) Kalman filters are designed to combine the outputs of two Kalman filters, each based on vehicle kinematics and bike models. To optimally combine the outputs of these two Kalman filter outputs, we compute the weighted probabilities of each output based on a probabilistic method that reflects each model feature in real time. The final estimated performance of the proposed IMM Kalman filter was confirmed based on the experimental results. In particular, comparisons of single Kalman filters and estimation accuracy are performed in detail. The main advantages of the proposed estimation algorithm are summarized as 1) optimality according to vehicle model combination and 2) accurate estimation of tire cornering stiffness.