Gravity-Constrained Map-Frame Initialization and Loop-Enhanced Cooperative Mapping for Air-Ground LiDAR SLAM
Year: 2026
Type: conference-paper
Abstract: Air-ground cooperative mapping combines the wide-area perception of UAVs and the dense near-ground observations of UGVs, but cross-platform initialization and common-view loop detection remain challenging due to viewpoint, height, and density differences. This paper proposes a Voxel-SLAM-based cooperative LiDAR mapping framework for UAVUGV systems. A gravity-guided air-ground initialization method is first introduced, where static IMU measurements and ground fitting provide gravity and height priors to reduce the initial alignment to an SE(2)-constrained voxelized GICP problem. A hybrid loop descriptor is then used to detect cross-platform loop constraints through global density retrieval and local triangular feature verification. The verified constraints are incorporated into global optimization for consistent map fusion. Experiments on a campus-garden dataset show that the proposed initialization improves air-ground map-frame alignment, while the loop-enhanced global optimization reduces the trajectory RMSE from 2.7703 m to 0.1631 m. These results demonstrate the feasibility and effectiveness of the proposed framework in the tested semi-structured outdoor environment, and its generalization to more diverse air-ground scenarios will be investigated in future work.
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204882983