Insufficient environmental information indoor localization of mecanum mobile platform using wheel-visual-inertial odometry

In autonomous driving of the mobile robot, the robot's current location should be identified first to plan and move a path to the target location. Accordingly, research on the robot's localization using GPS, 3D LiDAR, and Vision has been actively conducted. However, there is a limitation in that it is difficult to locate robots in indoor spaces where signals are disturbed by walls or ceilings, or in areas where sufficient environmental information cannot be obtained. This paper introduces the robot's position estimation method to overcome these environmental problems by using sensor fusion in an indoor tennis court. We propose a localization method that has low latency performance and high location accuracy through the use of Kalman filters to fuse data from wheel odometry and visual-inertial odometry. To evaluate its performance, this method was compared against wheel odometry, visual-inertial odometry, and LIO-SAM after the robot completed three rectangular paths. The resultant mean absolute errors in the x and y directions were 1.908 m and 0.707 m for wheel odometry, 1.169 m and 1.430 m for visual-inertial odometry, and 0.400 m and 0.383 m for LIO-SAM, respectively. In contrast, the wheel-visual-inertial odometry introduced in this study reported errors of 0.209 m and 0.103 m in the x and y directions, respectively, indicating superior accuracy compared to the other algorithms. This underscores the effectiveness of the proposed method in indoor environments where signals can be obstructed by walls or ceilings, or in areas lacking abundant environmental information.