Optical Flow-Based Pose Correction for Robust Robot Localization in Corridor Environments

Abstract

In this paper, we address the problem of robot localization using only LiDAR (light detection and ranging) in corridor environments. It is highly difficult to conduct robustly robot localization in corridor environments due to the lack of salient features for scan matching. To address these issues, this paper proposes an optical flow-based method to correct robot poses by preventing wrong localization result. We use the optical flow method to correct incorrect position estimates using the optical flow results when the localization results are going backwards, but the actual direction of robot movement as determined by the optical flow algorithm is forward. In addition, for more accurate position estimation, we combine the MonteCarlo localization (MCL) and LiDAR odometry localization results using a consensus filter. The proposed method was implemented in ROS (robot operating system) and showed reduced localization errors compared to robot localization results without the proposed method.