A Comparative Study: L1-Norm vs. L2-Norm; Point-to-Point vs. Point-to-Line Metric; Evolutionary Computation vs. Gradient Search

摘要

The study presented in this article compares the two most frequently used regularizations in the robotics area: L1-norm and L2-norm, for navigation purposes of an autonomous mobile platform in an inner environment with use of the 2D laser scanner. Sensorial data in a real environment are very often burdened by a noise, which unfavorably affects the classification process. Presented results show behavior of all tested algorithms under conditions in which the sensorial data are loaded by common types of the noise: moving objects, quantizing noise, artificially added noise with different types of characteristics that correspond to potentially real conditions. Basic navigation mechanisms presented here use methods of robust statistics and modern evolutionary optimizers. The methods selected in this study represent the two different types of metrics, commonly called point-to-point and point-to-line. The navigation algorithm that uses L1-norm regularization integrates several different evolutionary algorithms that occupy a position of very efficient optimizers, which, at the same time, do not cut down limits of usability of the tested methods. Correct working parameters settings of all used pose estimators play a key role in the robot pose and heading estimation, therefore, this article is extended by a description of several important working parameters and the way to use them.