Abstract EANA2026-69

Advancing the autonomous scientific decision-making capabilities of surface rovers is critical to maximizing the scientific return of upcoming planetary missions. We present an innovative autonomous mineral prospecting and path-planning software architecture designed to maximize astrobiological returns, reduce overall mission operational costs, and improve data collection efficiency for future planetary exploration missions, such as ExoMars. Our system integrates an onboard multispectral imaging pipeline, covering 0.25 to 3.5 µm spectral range, with a supervised machine learning classifier. This model was trained on reflectance spectra from the USGS spectral Library. By evaluating a 213-dimensional feature vector that preserves characteristic spectral shapes, the model currently achieves a significant classification accuracy of 87.5% across 105 mineral classes.

To directly bridge machine learning with astrobiological objectives, the classification pipeline computes a Mineral Relevance Score (MRS). The MRS applies a three-tier prioritization system to the detected mineral classes, explicitly weighted by the mineral’s relation to past liquid water activity, aqueous alterations, and its potential to preserve organic molecules or biosignatures. The MRS is divided into:

1. High-Importance Targets (MRS - HIT): Phyllosilicates, hydrated sulfates, hydrated silica, and carbonates, which are heavily prioritized as indicators of sustained habitable environments (mineral biomarkers);
2. Mid-Importance Targets (MRS - MIT): Sulfate evaporates (gypsum, jarosite, kieserite, basanite) and iron oxides, which serve as relevant indicators of past water and potential energy sources, despite highly acidic or saline environmental conditions;
3. Low-Importance Targets (MRS - LIT): Igneous silicates (olivine, pyroxene, feldspar) and perchlorates, where minerals lack interaction with liquid water or actively destroy potential organic signatures.

This prioritization framework directly maps to the expected Noachian stratigraphy of the ExoMars landing site at Oxia Planum, ensuring the system is heavily biased toward the region’s dominant clay-bearing plains and deltaic hydrated silica deposits.

Validated within a simulated operational environment, the MRS dynamically feeds into a higher-interest gradient path-planning algorithm optimized for execution on an On-Board Computer (OBC). When valuable (MRS - HIT) are detected, the software autonomously drives the rover to follow the specific mineral concentration gradient, creating a directional bias toward optimal aqueous-alteration zones. In the absence of target minerals, it executes an expanding search pattern until a significant spectral signature is triggered. A primary output of this autonomous phase is the generation of a comprehensive spatial maps detailing the detected minerals, their specific concentrations, and the corresponding MRS topography across the mission site.

Crucially, by executing these decisions entirely onboard, the system eliminates the operational bottleneck of Earth-Mars telemetry delays, drastically improving the speed and efficiency of scientific data collection while allowing for a leaner ground-control staffing model, ultimately reducing mission operational costs. Through translating spectral data into astrobiologically weighted concentration maps, this AI system can equip ExoMars or equivalent missions to autonomously investigate the most promising targets with more efficiency.