What the study found
The study found that STELLA, described as the first end-to-end crater-based navigation (CBN) pipeline for long-duration lunar mapping, can maintain metre-level position accuracy and sub-degree attitude accuracy on average in a lunar mapping setting.
Why the authors say this matters
The authors say this bridges a gap between CBN studies for powered descent and landing and the harder conditions of lunar mapping missions. They conclude that the results provide the first comprehensive assessment of CBN in a true lunar mapping setting and inform operational conditions to consider for future missions.
What the researchers tested
The researchers tested STELLA, which combines a mask R-CNN-based crater detector, a descriptor-less crater identification module, a robust perspective-n-crater pose solver, and a batch orbit determination back-end. They evaluated it on CRESENT+ and on CRESENT-365, a new public dataset designed to emulate a year-long lunar mapping mission with 15,283 rendered images from high-resolution digital elevation models, SPICE-derived Sun angles, and Moon motion.
What worked and what didn't
Experiments on CRESENT+ and CRESENT-365 showed that STELLA maintained metre-level position accuracy and sub-degree attitude accuracy on average across wide ranges of viewing angles, illumination conditions, and lunar latitudes. The abstract does not report specific failure cases or detailed performance breakdowns beyond noting the breadth of conditions tested.
What to keep in mind
The abstract does not give numerical error values, comparisons to other systems, or detailed limitations. It also does not describe which operational conditions were most challenging, only that the tested missions involved sparse, oblique imagery and varying illumination over long durations.
Key points
- STELLA is presented as the first end-to-end crater-based navigation pipeline for long-duration lunar mapping.
- On average, it achieved metre-level position accuracy and sub-degree attitude accuracy.
- The authors created CRESENT-365, a public dataset with 15,283 images to emulate a year-long lunar mapping mission.
- Tests covered wide ranges of viewing angles, illumination conditions, and lunar latitudes.
- The authors say the work provides the first comprehensive assessment of CBN in a true lunar mapping setting.
Disclosure
- Research title:
- AI crater navigation maintains accuracy for lunar mapping
- Authors:
- Sofia McLeod, Chee-Kheng Chng, Matthew Rodda, Tat-Jun Chin
- Institutions:
- South Australian Museum, The University of Adelaide
- Publication date:
- 2026-04-28
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.