Robust robot localization with visually adaptive consensus filters in dynamic corridor environments

This paper deals with the problem of robot localization in dynamic corridor environments. If a robot uses only a LiDAR (light detection and ranging) for its localization, the accuracy of robot localization degenerates as time goes due to the occlusions by moving people around a robot and the lack of scan features in corridors. This paper proposes a robust robot localization method with visually adaptive consensus filters (VACF) to solve the problem. The VACF consists of LiDAR odometry estimation, probabilistic localization, visual odometry estimation, optical flow recognition, object detection and adaptive consensus filters. To deal with long corridor environments, optical flow methods are used to correct the robot’s position. For robust localization in dynamic environments, object detection algorithm is used to detect dynamic objects, and localization algorithms are adaptively used as input to a consensus filter based on the number of dynamic objects detected. The VACF was tested in real-world experiments in dynamic corridor environments and showed better accuracy than other existing methods when compared to pre-determined ground truth points.