Robust mobile robot self-localization by soft sensor paradigm

The Mobile Robot Self-Localization is always a crucial aspect of the autonomous navigation task. The challenge of self-locating become complicated when the robot has sensors having low-level precision and accuracy. This work faces this aspect finding a solution by the using of the soft sensor paradigm. Various sources of information regarding the robot localisation are involved in a data fusion mechanism to get a more accurate estimation of the position of a mobile robot. Statistical considerations concerning the probability of a correct estimate for each source of information constitute the kernel of the soft sensor for the mobile robot self-localization. The soft sensor also computes the geometric transformations needed to correct all the different positions of the robot achieved by each source of information. Moreover, the paper reports an experiment of localization based on the combination of measures arising from a probabilistic approach (based on Adaptive Monte Carlo Localization) and the robot odometry. The proposed approach improves the accuracy of the autonomous navigation by means of a dynamic choice of the best available measure at any moment.