A Human-Robot Collaborative System for Robust Three-Dimensional Mapping

Map building is a fundamental problem in many robotic applications. Currently, most robots still lack sufficient high-level intelligence to achieve robust, efficient, and complete mapping of real world environments. In this paper, we develop a human-robot collaborative three-dimensional (3-D) mapping system based on a mobile robot platform equipped with a rotating RGB-D camera. This system introduces the robot's capability of quantitatively evaluating the mapping performance with human remote guidance. In this way, the robot is able to proactively cooperate with the human operator in real time for improved mapping performance. First, a Bayesian framework is proposed that fuses robot motion and visual features for regular 3-D mapping. Second, a binary hypothesis testing problem is formulated to evaluate the accuracy of camera pose estimation. When the estimated pose has small errors, the camera configuration is saved as a safe camera pose (SCP). When the estimated pose has large errors, a self-recovery mechanism is introduced allowing the robot to trace back to the last saved SCP. The proposed system is tested in different scenarios. The experimental results show that the system can run in real time and improve the accuracy and robustness of the mapping process.