Vehicle inertial parameter identification using Extended and unscented Kalman Filters

Abstract

Contemporary safety systems, such as obstacle avoidance and lane-keeping assistance, require good approximations of vehicle inertial properties, such as sprung mass and yaw moment of inertia, which can vary significantly based on the number of passengers, seating arrangement, and luggage. This paper demonstrates the implementation of two model-based parameter estimation algorithms, the Extended Kalman Filter (EKF) and the Unscented Kalman Filter (UKF), which are capable of working with a four degree of freedom, nonlinear vehicle model. While the EKF requires analytical linearization of the vehicle model at each step, the UKF approximates the parameter distribution with discrete sigma points and propagates them through the original nonlinear system. Simulation of a double lane change in CarSim illustrates the superior performance of the UKF for vehicle inertial parameter identification.