3D Semantic Mapping

3D semantic mapping aims to create detailed three-dimensional models of environments that are annotated with semantic labels, identifying objects and their locations. Current research emphasizes efficient and scalable methods, often employing Gaussian splatting or voxel-based representations, and incorporating techniques like hierarchical categorization and uncertainty-aware fusion to improve accuracy and reduce computational costs. This technology is crucial for advancing robotics, autonomous navigation, and applications requiring precise scene understanding, particularly in complex or unstructured environments like off-road terrains or underwater settings. Improved accuracy and real-time performance are key focuses, with ongoing efforts to address challenges like overconfidence in map estimations and efficient handling of large-scale datasets.