Planar-Feature Based 3D SLAM Using Randomized Sigma Point Kalman Filters

Abstract

In this study, a novel filtering method called Randomized Sigma Point Kalman Filter (RSPKF) is introduced for feature based 3D Simultaneous Localization and Mapping (SLAM). Conventional SLAM methods are mostly based on Extended Kalman Filters (EKF) for ‘mild’ nonlinear processes and Unscented KF (UKF) or Cubature KF (CKF) for ‘aggressive’ nonlinear processes. A critical problem of the existing filtering methods is that they lead to biased estimates of the state and measurement statistics. The main purpose of this study is to propose a new local filter, RSPKF, based on stochastic integration rules providing an unbiased estimate of an integral for feature based SLAM. The simulation based on point features in 2D and experimental results based on planar features in 3D show that the RSPKF based SLAM method provides more accurate results than the traditional methods.