Abstract
Aiming to accomplish automatic and real-time three-dimensional mapping in both indoor and outdoor scenes, a low-cost wheeled robot-borne laser scanning system is proposed in this paper. The system includes a laser scanner, an inertial measurement unit, a modified turtlebot3 two-wheel differential chassis and etc. To achieve a globally consistent map, the system performs global trajectory optimization after detecting the loop closure. Experiments are undertaken in two typical indoor/outdoor scenes that is an underground car park and a road environment in the campus of Wuhan University. The point clouds acquired by the proposed system are quantitatively evaluated by comparing the derived point clouds with the ground truth data collected by a RIEGL VZ 400 laser scanner. The results present an accuracy of 90% points below 0.1 meter error in the tested scene, showing that its applicability and potential in indoor and mapping applications.
Introduction
Traditional Mobile Mapping Systems (MMS) are the gold standard for creating 3D digital twins, but they face significant hurdles in cost and environmental constraints. High-end commercial systems often exceed $300,000, and their reliance on GNSS (GPS) for positioning makes them ineffective in "urban canyons," under heavy tree canopies, or inside underground structures. This research addresses the need for a low-cost solution capable of navigating GPS-denied areas autonomously while maintaining centimeter-level precision.
System Configuration and Methodology
The proposed system is built upon a modified Turtlebot3, a two-wheeled differential drive robot. The hardware stack integrates mid-range components to balance cost and performance:
- LiDAR: A Velodyne VLP-16 (Puck) for 3D point cloud acquisition.
- IMU: An Xsens MTi-300 AHRS for inertial data.
- Compute: An Intel NUC i7 mini-PC for real-time processing.
The software pipeline utilizes LeGO-LOAM (Lightweight and Ground-Optimized Lidar Odometry and Mapping), which separates the estimation problem into high-frequency odometry and lower-frequency mapping tasks. To mitigate the drift inherent in SLAM systems, the authors implemented global trajectory optimization. Upon detecting a loop closure, the system employs a factor graph to retroactively correct the trajectory, ensuring global consistency.
Experimental Results
The system was evaluated in two real-world scenarios at Wuhan University: a $1200m^2$ underground parking lot (GPS-denied) and a campus road environment with heavy tree cover (weak GNSS). Results were validated against ground truth data from a high-precision RIEGL VZ 400 Terrestrial Laser Scanner.
Quantitative analysis in the indoor parking lot showed that 90.4% of the points were within 0.1 meters of the ground truth. Qualitatively, the system successfully captured fine details of tree trunks, parked cars, and building facades in real-time.
Conclusion
This work demonstrates that professional-grade 3D mapping results can be achieved using mid-range hardware through sophisticated SLAM algorithms and global optimization. While the system performs well in moderate-sized areas, future work will focus on scaling to larger industrial complexes and integrating cameras to improve robustness in featureless environments.
Summary updated