A Novel Approach for Lidar-Based Robot Localization in a Scale-Drifted Map Constructed Using Monocular SLAM

Area of Science:

• Robotics
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Monocular SLAM systems suffer from scale ambiguity and drift, hindering metric consistency.
• This inconsistency poses challenges for robots using different sensors, like laser range finders (LRF), to localize in pre-existing maps.

Purpose of the Study:

• To develop a 2D-LRF-based localization algorithm for robots.
• To enable robots to self-localize and simultaneously resolve the scale of a map generated by monocular SLAM.
• To address the metric inconsistency issue in cross-sensor mapping and localization.

Main Methods:

• 2D structures are extracted from 3D point cloud maps generated by visual SLAM.
• A modified Monte Carlo Localization (MCL) approach is employed to estimate the robot's pose and the map's relative scale.
• The algorithm aligns 2D LRF data with the extracted map structures.

Main Results:

• The proposed system demonstrates effective real-time global localization of robots.
• Successful localization is achieved even when operating within a significantly drifted map.
• Experimental validation on a public dataset and a real-world scenario confirms the method's efficacy.

Conclusions:

• The developed algorithm successfully overcomes scale ambiguity in monocular SLAM-generated maps.
• It provides a robust solution for LRF-equipped robots to localize accurately in metric-inconsistent environments.
• The method offers real-time performance and resilience to map drift, enhancing robot navigation capabilities.