CrossMaps: Confidence-Aware Open-Vocabulary Semantic Mapping for Rover Navigation
Researchers have developed CrossMaps, a novel pipeline for real-time, confidence-aware semantic mapping designed for rover navigation. This system integrates multi-scale CLIP embeddings with a dual-memory architecture (STM and LTM) to create language-queryable maps from RGB-D data. CrossMaps processes visual observations, fuses them with confidence cues, and promotes coherent semantic landmarks to long-term memory, enabling natural language queries to guide rover movement. AI
IMPACT Enables more intuitive and flexible rover navigation through natural language commands.