Spatial Commonsense

Spatial commonsense research focuses on enabling artificial intelligence systems to understand and reason about the spatial relationships between objects in a scene, mirroring human intuitive understanding. Current efforts concentrate on developing robust scene representations, often employing graph-based models enriched with commonsense knowledge from external knowledge bases, and leveraging both visual and textual data to train models capable of accurate object localization and spatial reasoning tasks. This research is crucial for advancing robotics, particularly in areas like object search and navigation in dynamic environments, and for improving the performance of large language models in tasks requiring spatial understanding.