Vision-based mobile robot navigation using active learning concept

How to find out the safety region and moving direction is an important research issue in autonomous mobile robot navigation. It is well known that color is one of the prominent path features. We make use of the features to provide useful information for robot navigation in our work. We construct path's Gaussian Mixture Model (GMM) from image data. Unfortunately the result of path recognition shows that some outlier should be recognized as parts of path. To deal with this problem, we induct active learning concept to construct extra model for these outlier. Experimental results show that our approach increases the accuracy of path recognition. Finally the result of path recognition is used to make decision for motor command generation to control the mobile robot. The performance of the proposed approach is verified in real workspace, demonstrating its superiority.