Two-Stage Kalman Filtering for Indoor Localization of Omnidirectional Robots

Abstract: In this study a sensor fusion technique was developed for indoor localization of omnidirectional mobile robots. The proposed sensor fusion method combines the measurements made by an indoor localization system (e.g. ultrasound based localization) with the measurements that comes from an IMU (Inertial Measurement Unit). It was taken into consideration that the measurements made by the ultrasonic sensor are available with lower measurement rate than the IMU measurements. To tackle this problem a two-stage Kalman filter was developed. The first stage estimates the orientation of the robot at the measurement rate of the IMU using a linear Kalman filter. The second stage estimates the position and velocity of the robot. The prediction of the robot’s position is performed based on the robot model at the update rate of the IMU. The correction phase is performed at the update rate of the external localization system using nonlinear Kalman filtering techniques. All the measurements were considered noisy and in the case of the IMU the measurement biases were also taken into consideration. For the implementation of the nonlinear Kalman filter, a discretized nonlinear omnidirectional robot model was developed. It was necessary as the classic unicycle model cannot be directly applied in the case of the omnidirectional robot. Simulation measurements were performed to evaluate the performances of the proposed two-stage filter. For the implementation of the second stage, two approaches were tested namely the Extended Kalman Filter and the Unscented Kalman Filter.