3D Scene Graph

3D scene graphs represent environments as layered graphs, with nodes representing objects and places, and edges depicting their relationships, aiming to provide robots with a richer understanding of their surroundings than traditional geometric maps. Current research focuses on developing efficient algorithms for scene graph construction and optimization, often employing graph neural networks and large language models to integrate semantic and geometric information, and to handle open-vocabulary objects and relationships. This research is significant for advancing robotics, particularly in areas like navigation, task planning, and human-robot interaction, by enabling more robust and adaptable autonomous systems.