Development of swarm robot system using visual-inertial-UWB sensor fusion in an indoor environment with sparse feature points

Abstract

The technology for estimating and correcting a robot's pose is still a challenging field in the Global Navigation Satellite System (GNSS) denied environment. In particular, there are many difficulties in estimating the pose of robots based on a mono camera in an indoor environment with sparse feature points. In such an environment, we propose an algorithm that performs robust pose estimation through sensor fusion of individual robots and revises once again when the same place is revisited. In more detail, fusing the Ultra-WideBand (UWB) sensor to the Visual Inertial Odometry (VIO) algorithm affects the initialization and scale correction of the existing VIO algorithm for a more robust result. In addition, a cost function was designed in consideration of the UWB signal characteristics, which are less reliable as the distance between the UWB sensors increases. When the robot revisits, the pose correction process called loop closure is performed. We propose a correction method using up to a Visual-Inertial-UWB sensor. The aforementioned method of pose estimation and correction of individual robots is applied to the swarm robot system, which is proposed. The proposed swarm robot system transmits the estimated pose of individual robots to ground station capable of high-performance computation. And ground station aligns the pose of each robot based on the global frame by the pose correction and reference frame of the swarm robot system. As a result, a more robust pose was obtained than the conventional VIO algorithm used for individual robots. Also, a more robust pose was obtained in the algorithm extended to the swarm robot system. By applying the proposed method to the swarm robot system, we can expect to create various missions and global maps in the future.