ALPER: Vision based absolute localization for planetary exploration rovers - Statistical analysis of complementary approaches

Future surface planetary missions, such as sample returns or base construction, require advanced Guidance, Navigation, and Control systems for long-distance autonomous operations. Current systems depend on relative localization, which accumulates error over extended distances. Absolute localization algorithms, akin to GNSS on Earth, are crucial for accurate, independent pose corrections.

Focusing initially on the Martian context and the availability of satellite imagery with 0.3 m/px resolution, the ALPER project developed three absolute localization algorithms: CM (Constellation Matching) using rock detection, DICOR (Dense Image Co-registration) aligning orthomosaics with orbital images, and SkyM (Skyline Matching) for horizon skyline comparison. These methods build upon landmark matching, dense image matching, and horizon data alignment techniques, introducing improvements in both the data processing and matching stages.

CM and DICOR were evaluated through Monte Carlo simulations and field trials in the Bardenas Reales, achieving localization accuracy below 0.6m under nominal conditions, and within the 1.25m criteria in 94 %–99 % of the cases, even with initial offsets of up to 20 m. The algorithms demonstrated complementary strengths: CM was effective across varying illumination conditions, while DICOR proved robust to rock density. These solutions have shown robustness, readiness, and high effectiveness for future planetary missions, reaching TRL6 maturity. SkyM obtained heading errors under 2° and position errors around 5 m, indicating promising potential for further development to increase its maturity level.

Preliminary results are also presented regarding the evaluation and adaptation of these localization methodologies to the lunar South Pole context, characterized by harsh illumination conditions. Tests are realized on a simulated realistic lunar dataset, and the method’s performances are compared to those obtained on the Martian context. The study then introduces advancements to their design, with the key improvement being the detection and use of craters as landmarks. First functional tests demonstrate that this enhancement extends the method’s operational domain, increasing the amount of estimations and overall accuracy.