CrossMaps: Confidence-Aware Open-Vocabulary Semantic Mapping for Rover Navigation

CrossMaps is a newly introduced real-time confidence-aware open-vocabulary semantic mapping pipeline designed for rover navigation, leveraging RGB-D data. It utilizes a dual-memory architecture with Short-Term Memory (STM) for aggregating noisy observations and Long-Term Memory (LTM) for maintaining persistent semantic landmarks, integrating multi-scale CLIP embeddings for enhanced semantic understanding. This development is significant for practitioners as it enables the creation of language-queryable maps that improve navigation under partial observability, facilitating better decision-making in autonomous systems.