Climb-Odom: A robust and low-drift RGB-D inertial odometry with surface continuity constraints for climbing robots on freeform surface

Low-drift localization and reliable yaw orientation estimation remain challenging for visual-inertial odometry (VIO) without loop closure, especially for climbing robots in weak-featured environments. In this article, we proposed a robust and low-drift RGB-D inertial odometry named Climb-Odom for climbing robots on large freeform surfaces. The odometry consists of a real-time 6-DOF (degree of freedom) pose estimator and an optimized point cloud map. The estimator incorporates surface continuity constraints in addition to visual and inertial constraints to realize low-drift localization. The point cloud map is reconstructed rectifying translation errors in surface normal and rotation errors around gravity by map optimization and merging. On the low-textured and freeform challenging surface, Climb-Odom still shows robust and accurate localization performance and effectively reconstructs the traveled area with high quality. Experiments with the climbing robot on a real wind turbine blade demonstrate that Climb-Odom outperforms state-of-the-art RGB-D inertial odometry and SLAM systems. For the zig-zag trajectories covered by the climbing robot, the localization root mean square error (RMSE) of Climb-Odom are less than 0.42% in all datasets while aligned by the origins. The RMSE of iterative closest point (ICP) result between the largest reconstructed map and the reference point cloud is less than 7 mm within approximately 5.6 m 2 . • Surface continuity constraints based on continuous assumption of freeform surfaces. • Data association method for featureless depth measurement. • Map optimization module for merging featureless marginalized point clouds. • Low-drift and real-time RGB-D inertial odometry for climbing robots. • It outperforms other state-of-the-art RGB-D inertial odometry or SLAM systems.