A pose graph-based localization system for long-term navigation in CAD floor plans

• Published in: Robotics and autonomous systems Vol. 112; pp. 84 - 97
• Main Authors: Boniardi, Federico, Caselitz, Tim, Kümmerle, Rainer, Burgard, Wolfram
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.02.2019
• Subjects: Adaptive systems; Localization; Mapping; Mobile robotics; SLAM; Adaptive systems; Mapping; SLAM; Localization; Mobile robotics
• ISSN: 0921-8890; 1872-793X
• DOI: 10.1016/j.robot.2018.11.003

Accurate localization is an essential technology for flexible automation. Industrial applications require mobile platforms to be precisely localized in complex environments, often subject to continuous changes and reconfiguration. Most of the approaches use precomputed maps both for localization and for interfacing robots with workers and operators. This results in increased deployment time and costs as mapping experts are required to setup the robotic systems in factory facilities. Moreover, such maps need to be updated whenever significant changes in the environment occur in order to be usable within commanding tools. To overcome those limitations, in this work we present a robust and highly accurate method for long-term LiDAR-based indoor localization that uses CAD-based architectural floor plans. The system leverages a combination of graph-based mapping techniques and Bayes filtering to maintain a sparse and up-to-date globally consistent map that represents the latest configuration of the environment. This map is aligned to the CAD drawing using prior constraints and is exploited for relative localization, thus allowing the robot to estimate its current pose with respect to the global reference frame of the floor plan. Furthermore, the map helps in limiting the disturbances caused by structures and clutter not represented in the drawing. Several long-term experiments in changing real-world environments show that our system outperforms common state-of-the-art localization methods in terms of accuracy and robustness while remaining memory and computationally efficient. •Build a pose graph online using trajectory priors to align it to the floor plan.•Switch to pure localization whenever the environment is sufficiently mapped.•Estimate the robot pose using relative localization with respect to the pose graph.•Maintain an up-to-date pose graph in changing environments using Bayes filtering.•Bound memory consumption and runtime to enable long-term operation.