Proactive 3D Robot Mapping in Environments with Sparse Features

Abstract

3D map building can aid robots to accomplish high-level tasks. Using an inexpensive RGB-D camera, a 3D map can be built by estimating the camera pose using visual features. However, the mapping will easily fail if there lack a sufficient number of features. In this paper, a proactive 3D mapping framework is proposed using a mobile robot platform equipped with an RGB-D camera and a projector. Both the camera and the projector are mounted on pan-tilt units controlled by servo motors. With the motion of the camera pan-tilt unit and the movement of the robot, a binary hypothesis testing problem is modeled to evaluate the estimation accuracy of the camera pose. A pattern is generated by the projector to increase the number of features when the pose estimation has large errors based on the real-time evaluation. The experimental results show that the proposed approach improves the mapping performance in an indoor environment with sparse features.