SLAM Based on Double Layer Unscented Kalman Filter

The unscented Kalman filter based SLAM algorithm (UKF-SLAM) has the problem of the inaccurate estimation and the system instability. A novel SLAM algorithm based on double layer unscented Kalman filter (DLUKF-SLAM) is presented in this paper to overcome the important drawbacks of the previous frameworks. The proposed algorithm first uses a set of weighted sampling points to represent the prior distribution of states, and then the inner layer UKF-SLAM is performed to achieve the states prediction, finally, the estimations of the robot pose and landmark position at each time are obtained by the update mechanism of the outer layer UKF-SLAM. Simulation results demonstrate the superiority of the DLUKF-SLAM algorithm in the aspect of estimation accuracy and stability, compared with the EKF-SLAM algorithm and the UKF-SLAM algorithm.