2020/01/28 by Milad Ramezani, Ramezani, Milad, Georgi Tinchev +5 · 2 citations
Engineering · #FOS: Computer and information sciences #Machine Learning (cs.LG) #Robot Manipulation and Learning #Robotic Locomotion and Control #Robotics (cs.RO) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.2001.10249
openalex publication_date 2020/01/28 · openalex created_date 2021/11/08 · openalex updated_date 2026/07/28
In this paper, we present a factor-graph LiDAR-SLAM system which incorporates\na state-of-the-art deeply learned feature-based loop closure detector to enable\na legged robot to localize and map in industrial environments. These facilities\ncan be badly lit and comprised of indistinct metallic structures, thus our\nsystem uses only LiDAR sensing and was developed to run on the quadruped\nrobot's navigation PC. Point clouds are accumulated using an inertial-kinematic\nstate estimator before being aligned using ICP registration. To close loops we\nuse a loop proposal mechanism which matches individual segments between clouds.\nWe trained a descriptor offline to match these segments. The efficiency of our\nmethod comes from carefully designing the network architecture to minimize the\nnumber of parameters such that this deep learning method can be deployed in\nreal-time using only the CPU of a legged robot, a major contribution of this\nwork. The set of odometry and loop closure factors are updated using pose graph\noptimization. Finally we present an efficient risk alignment prediction method\nwhich verifies the reliability of the registrations. Experimental results at an\nindustrial facility demonstrated the robustness and flexibility of our system,\nincluding autonomous following paths derived from the SLAM map.\n