UMP Institutional Repository

Smoothing techniques and a spherical simplex unscented transformation in solving a SLAM problem

Saifudin, Razali (2012) Smoothing techniques and a spherical simplex unscented transformation in solving a SLAM problem. PhD thesis, Okayama University.


Download (1MB)


This thesis focuses on the use of unscented transformation method to solve a simultaneous localization and mapping (SLAM) problem. SLAM is the process by which a mobile robot can build a map of an environment and at the same time use this map to compute its location.It can be performed by storing landmarks in a map when they are observed by the robot sensors, using the robot pose estimate to determine the landmark locations,while at the same time, using these landmarks to improve the robot pose estimate. Since the landmarks are repeatedly reobserved, their locations become increasingly certain and the map converges, eventually acquiring the rigidity of a priori map.Many solutions to the SLAM problem are focused on the filtering approaches such as the use of the extended Kalman filter (EKF),the unscented Kalman filter (UKF),the particle filter (PF) as well as their variations.However,the smoothing approach has received less attention in solving this problem.Therefore,this thesis presented a smoothing approach to solve the SLAM problem,by the implementation of Rauch-Tung-Striebel smoother.In the beginning,a linearization approach has been applied to the Rauch-Tung-Striebel smoother.This smoother named as extended Rauch-Tung-Striebel smoother (ERTSS).The performances of this smoother is better compared to the standard EKF.In order to minimize errors in the nonlinear estimation,this thesis utilizes the benefit of an unscented transformation over linearization in the ERTSS.In the unscented transformation,the state distribution is represented using a minimal set of carefully chosen sample points,called sigma points.These sigma points are propagated through the true nonlinear function,without any approximation.In addition, the difficulty of Jacobian matrix calculation, which is used in the EKF, is not required in this method. This transformation method is applied to the Rauch-Tung-Striebel smoother to obtain the unscented Rauch-Tung-Striebel smoother (URTSS).The performance of the URTSS is evaluated and compared to the similar filtering method, the UKF.The result shows that the URTSS gives lower errors in solving the SLAM problem, compared to the errors produced by the UKF.This thesis also investigates other paradigm of solving a SLAM problem known as a FastSLAM approach. In this framework, the approximation used in the standard FastSLAM is replaced by the unscented transformation,in which it is called the unscented FastSLAM (UFastSLAM).The proposed method is evaluated and its performance is compared to the standard FastSLAM.It is shown that, the UIFastSLAM gives better result in solving the SLAM problem.In addition,a new sampling technique, which is called a spherical simplex unscented FastSLAM (SSUFastSLAM),is presented.This new sampling technique uses less number of sigma points, compared to the standard one used in the UFastSLAM.For that reason, the computational cost is reduced without giving any effect on its performance,which is proved by the simulation result.

Item Type: Thesis (PhD)
Uncontrolled Keywords: Mobile robots; Mapping; Localization theory;
Subjects: T Technology > TJ Mechanical engineering and machinery
Depositing User: Shamsor Masra Othman
Date Deposited: 12 Nov 2013 02:57
Last Modified: 23 Aug 2017 07:25
Download Statistic: View Download Statistics

Actions (login required)

View Item View Item