Bearing-only unscented smoothers for a visual SLAM

In the conventional SLAM problem, a laser range scanner that can obtain relative angle and distance between a robot and a landmark is used as an external sensor. However, such a sensor is expensive, so that a bearing-only SLAM is studied recently by using a cheap CCD camera. For such a bearing-only SLAM problem, there is a problem that an objective environment has to be measured at multiple distinct poses of the robot, and since there is a lack of information it is also called to be not easy for assuring that the self-position of the robot and the landmarks are estimated with a high accuracy. In this paper, we focus on unscented smoothers that can improve the estimation accuracy for a general SLAM using unscented Kalman filters and apply it to design a bearing-only unscented smoother for a visual SLAM problem. Simulations are presented to check the usefulness of the proposed method.