Robust SLAM: SLAM base on square root unscented Kalman filter

摘要

Simultaneous localization and mapping (SLAM) is concerned to be the key point to realize the real autonomy of mobile robot. Extended Kalman filter (EKF) is a popular choice to solve SLAM problem. However, EKF-SLAM is vulnerable to linearization errors, which can cause poor performance and inconsistency. UKF-SLAM is proposed to solve these problems. However, the conventional UKF-SLAM is based on the assumption that the system is exactly known, and their disturbances are stationary Gaussian noises with known statistics, which these assumptions might not hold in many real applications. In this work, in order to avoid these shortcomings and to increase the accuracy of the estimation, the robust SLAM (RSLAM) is introduced. RSLAM is based on square root unscented Kalman filter which is applicable for nonlinear system with non-Gaussian noises. The proposed method does not require the knowledge of the noise distributions, and the noises can be non-Gaussian, so it is more flexible and has less limitation in real applications. In addition, the RSLAM has consistently increased numerical stability comparing to UKF-SLAM because all resulting covariance matrices are guaranteed to stay semi-positive definite. To get more acceptable performance, the parameters of RSLAM (including the covariance process noise Q and the covariance measurement noise R) are tuned by using an adaptive neuro-fuzzy inference system. The superior performance of RSLAM over other UKF-SLAM is validated by Monte Carlo simulation. The results show improvement in the accuracy and consistency of the state estimates.