Object Map Building on Various Terrains for a Wheeled Mobile Robot

This paper presents an objects-based topological mapping algorithm on different floors with various objects using a wheeled mobile robot. The extended Kalman filter (EKF) with adaptive measurement noise according to the terrain type is proposed to estimate the position of the robot. If an infrared distance sensor detects an object, the robot moves around the object to obtain the shape information. The row-wise max-pooling with a convolutional neural network (CNN) is proposed to classify objects regardless of the starting position of the observation. Finally, the object map consisting of nodes and edges generated from the classified objects and the distance between objects. Experimental results showed that the proposed algorithm could improve an accuracy of position estimation of the robot and efficiently generated the object map on various terrains.