**5. Discussion**

The simulations and experiments compiled for compare different weighting and gating approaches for a PF SLAM provide consistent results that justify the implementation of the described stratification algorithm. First of all, its accuracy, measured in terms of RMSE, is superior. Secondly, the robustness of the algorithm is demonstrated by its having the highest percentage of successful runs. Last but not least, the amount of resampling procedures performed in different variants suggests that the method of data processing, in the stratified approach, provided the largest amount of information, as it led to more distinct differences in sample weights and more frequent resampling.

As mentioned earlier, the method of real-world data registration provided no groundtruth trajectory. This can be considered an additional source of uncertainty, in terms of accurate RMSE calculation, however, its impact should not be overestimated, because the INS/GNSS trajectory was available. Still, the removal of such an uncertainty can be pointed out as a future research direction, and we plan to implement our algorithm on a platform which will be capable of performing real-time kinematic (RTK) surveying. This will allow centimeter-level accuracy of positioning in providing reference trajectories.
