Visual SLAM incorporating wheel odometer for indoor robots

Abstract

An approach for incorporation of a wheel odometer in monocular visual SLAM (Simultaneous Localization and Mapping) for indoor robots is proposed in this paper. It is based on nonlinear optimization parameterized in 6DoF (Degrees-of-Freedom) manifold SE(3), and constrains the robot poses to be in 3DoF space. Furthermore, to deal with the problem that in complicated indoor environment where the tracking of visual SLAM may easily get lost in some scenes with hardly any features, the approach creates new maps when new features are available and thus continuing estimating poses and feature positions to improve the robustness. The odometer can provide relative poses between maps, and when image overlaps are detected, the maps can be merged to a consistent one. Experiments are conducted on an indoor robot platform equipped with an upward-looking camera, and experimental results show the effectiveness of the proposed approach.