Non-odometry SLAM and Effect of Feature Space Parametrization on its Covariance Convergence

This paper is aimed at non-odometry feature-based variant of SLAM (Simultaneous Localisation and Mapping) algorithm. The main focus has been given to processing of linearly constraint map landmarks observations. The observation model was simplified to linear and with this assumption is given derivation of three different variants of applicable SLAM algorithm - first variant which doesn't take into consideration any map feature constraints, second variant which optimally process data under linear constraints assumption and third reduced variant which is designed to slightly suboptimal processing with significantly lower computational complexity. The performance of described methods was tested on synthetic data and results of simulations are presented at the end of this paper.