TinySLAM [1] is one of the most simple SLAM methods but the original implementation [2] is based on the specific robot model and provided as the ad-hoc application. Its key feature is simplicity of implementation and configuration at cost of accuracy (as our tests shown). Some changes were made in the original algorithm in order to minimize an error of estimated trajectories. The introduced model of cell leads to an error decrease on almost all tested indoor sequences. The proposed dynamic probability estimator improves usage of coarse-grained maps when memory efficiency is more desirable than accuracy. Obtained quantitative measurements justify the changes made in tinySLAM in case the method is used in a relatively small environment.