Sensor fusion of delay and non-delay signal using Kalman Filter with moving covariance

Abstract

A movement of the omni-directional mobile robot is deflected by a slippage of its wheels, measurement such as acceleration, angular velocity and computer vision should be applied to compensate error from slippage of each wheel. Each sensor has advantages and drawbacks. For example, computer vision(CV) can measure the absolute position and the angle of the robot but it requires much time to process images which causes timing delay. On the other hand, inertial sensors such as accelerometers and gyroscope can swiftly response to its movement. Unfortunately, the position and the angle are obtained by integrating the measured signal which causes accumulated error of estimating the position and angle. In this paper, Kalman Filter is applied to implement fusion of the delay and non-delay data. When a delay signal is available, a classical method, which filter delay signal is re-performing Kalman operation at every step from the time of measured delay signal to current time. Therefore, we proposed method that only need to update the stored covariance between two different time instants. With less computational cost comparing to the classical method and the uniformity of the computation in every iteration, the efficiency of Kalman filter remains the same. Also, this method can be applied to soft-nonlinear case by utilizing Extended Kalman filter.