Topological local-metric framework for mobile robots navigation: a long term perspective

Publication Type:
Journal Article
Citation:
Autonomous Robots, 2019, 43 (1), pp. 197 - 211
Issue Date:
2019-01-31
Filename Description Size
10.1007s10514-018-9724-7 am.pdfAccepted Manuscript Version8.36 MB
Adobe PDF
Full metadata record
© 2018, Springer Science+Business Media, LLC, part of Springer Nature. Long term mapping and localization are the primary components for mobile robots in real world application deployment, of which the crucial challenge is the robustness and stability. In this paper, we introduce a topological local-metric framework (TLF), aiming at dealing with environmental changes, erroneous measurements and achieving constant complexity. TLF organizes the sensor data collected by the robot in a topological graph, of which the geometry is only encoded in the edge, i.e. the relative poses between adjacent nodes, relaxing the global consistency to local consistency. Therefore the TLF is more robust to unavoidable erroneous measurements from sensor information matching since the error is constrained in the local. Based on TLF, as there is no global coordinate, we further propose the localization and navigation algorithms by switching across multiple local metric coordinates. Besides, a lifelong memorizing mechanism is presented to memorize the environmental changes in the TLF with constant complexity, as no global optimization is required. In experiments, the framework and algorithms are evaluated on 21-session data collected by stereo cameras, which are sensitive to illumination, and compared with the state-of-art global consistent framework. The results demonstrate that TLF can achieve similar localization accuracy with that from global consistent framework, but brings higher robustness with lower cost. The localization performance can also be improved from sessions because of the memorizing mechanism. Finally, equipped with TLF, the robot navigates itself in a 1 km session autonomously.
Please use this identifier to cite or link to this item: