Feature-Based Scanning Lidar-Inertial Odometry using Factor Graph Optimization

—Localization of a mobile robot in the absence of anabsolute position sensor often relies on techniques such as visualor lidar-inertial odometry. While lidar has many advantages,the most capable sensors use scanning mechanisms, leadingto motion-distorted scans. Previous strategies used to accountfor robot motion when performing state estimation and outlierrejection have drawbacks for use on highly dynamic, resourceconstrained robots such as spacecraft during descent and landing.In this paper we develop a novel probabilistic factor for theinclusion of scanning lidar features, and an accompanying outlierrejection methodology. By using well-established, efficient featuretracking techniques, our image processing front end is bothreliable and amenable to FPGA implementation, both of whichare critical for operation on a spacecraft. We demonstrate ourtechnique on datasets from simulated planetary descent andlanding, and aggressive real-world handheld motion. The resultsshow that our system can be used to perform accurate lidarinertial odometry, even in highly dynamic scenarios.