SAME: Ground-Air Collaborative Semantic Active Mapping and Exploration

Abstract

The integration of ground and aerial robots, known as the Ground-Air collaborative system, has the potential to handle mapping and navigation tasks in complex environments by utilizing the advantages of multiple perspectives and high maneuverability. The significant differences in the visual fields and angles of UAV and UGV pose a challenge in collaboratively generating a consistently styled map to guide exploration. This paper proposes a method for representing the environment using semantic Octree units as a basis, which are fused to generate a consistent global map. We are the first to connect tightly Ground-Air collaborative mapping and navigation through semantics-centered elements. Through the exploration, we combine UAV's extensive visual coverage with UGV's close-range, precise observations to achieve a multi-layered reconstruction of the scene. The 2D semantic map generated through 3D map projection provides information for path planning, creating a positive feedback loop between high-quality mapping and autonomous exploration. Merging RGB images, depth point clouds, and semantic data, the UAV and UGV independently construct local Octomap maps, which are then merged into a cohesive global map through network communication. Through both simulation and real-world verification, this semantic-centred Ground-Air collaborative approach enhances both the precision of mapping and the efficiency of exploration.