Enhancement of Robot Navigation Systems via Human Pointing Instruction

Abstract

The determination of destinations and paths is a critical aspect of robot navigation. In recent years, research has been conducted on methods to provide robots with realistic destinations for use in practical situations. One such general method is a spatial search based on limited information such as the name of the target object or a photograph of the destination. However, if the rough location of the destination is known from the viewpoint of an instructor, and can be communicated to the robot, then it is possible to reduce the search load on the robot. Therefore, in this study, we propose a new task that integrates the pointed instructions of an operator using and augmented reality device into the robot's search route. However, the instructor may not be able to point directly to the route if it is within a blind spot. Thus, we propose a combined search method; where the pointing path is represented as a set of vectors rather than a sequence of points and used to generate a potential field which is integrated with an existing traditional robot navigation method. In this study, we evaluate the proposed method in an experiment using the 3D simulator Habitat, and show that the proposed method outperforms existing methods under the pointing path information.