Researchers have developed Star-Fusion, a novel multi-modal transformer architecture designed for precise celestial orientation determination in spacecraft navigation. This approach reframes the problem as a discrete topological classification task, utilizing spherical K-Means clustering to manage the complexities of the celestial sphere's non-Euclidean topology. The architecture integrates a SwinV2-Tiny transformer for feature extraction, a convolutional heatmap branch for spatial grounding, and an MLP for geometric anchoring. Experimental results show Star-Fusion achieving 93.4% Top-1 accuracy with an inference latency of 18.4 ms, making it suitable for real-time onboard satellite applications. AI
Summary written by gemini-2.5-flash-lite from 3 sources. How we write summaries →
IMPACT This new architecture could improve the accuracy and efficiency of autonomous spacecraft navigation systems.
RANK_REASON This is a research paper describing a new model architecture for a specific technical problem.