In this paper, we propose a novel airplane centric localization approach for robust and accurate robot 3D localization with respect to an airplane. For maintaining the coordinate reference on the airplane while its parking pose may be changed at every inspection/maintenance instance, we maintain only the airplane as the reference point cloud map, while excluding all information of its surrounding. Here, to prevent the chances of lidar scan points from objects near the real airplane to have an incorrect match with the reference airplane map, we implement a novel online filtering approach. This filter leverages on the previous robot pose estimate in map coordinate frame and the robot motion dynamics to perform a fast and efficient extraction of potential airplane points from the latest lidar scan. This approach also helps remove the necessity of learning specific airplane characteristics and thus, increases the easy application scope to different airplane models. Lastly, an enhanced map matching approach helps compensate for the lower amount of extracted/available sensor data by leveraging on high map density. The proposed localization approach is tested extensively in physical experiments and, the results confirm significant improvement in localization robustness and accuracy, compared to other state-of-the-art technologies.
This research is supported by core funding from: A*STAR SERC AME
Grant Reference no. : A18B2a0082