Developing Flexible Algorithms to Optimize Drive Paths for the ExoMars Rosalind Franklin Rover

In 2030, the ExoMars Rosalind Franklin Mission Rover (RFM) is scheduled to land at Oxia Planum, Mars, to search for the chemical building blocks of life [1]. The mission’s success depends not only on the rover’s scientific payload, but also on RFM’s ability to safely and efficiently traverse the Martian terrain: what terrains are safe to drive across; what terrains or features on the landscape are potential mobility hazards; and how efficiently can the rover make it from one point to another?

Extensive work has gone into characterizing the landing site at Oxia Planum including the creation of high-resolution digital elevation models [DEMs; e.g. 2], high fidelity geologic [3] and mineralogic mapping [e.g. 4], and machine-learning assisted landscape classifications [5,6]. Additionally, many studies have characterized the wider Oxia region, identifying widespread evidence for ancient fluvial [e.g. 7, 8] alteration, as well as modern aeolian reworking of the surface [9]. RFM engineers and mission scientists will use this scholarship, as well as in situ images and DEMs to get the rover from one location in Oxia Planum to another.

During this pre-launch phase of the mission, we were curious to test whether we could automate the creation of rover traversability paths between two arbitrary points at Oxia Planum in a geographic information system (GIS). Specifically, we wanted to answer three simple questions: (1) what is the safest path from point A to point B, (2) how quickly can we traverse that distance, and (3) therefore, how many driving sols are needed?  

First, we compared NOAH-H (The Novelty and Anomaly Hunter – HiRISE [5]) deep learning terrain classifications at Jezero Crater [10] to Oxia Planum [5, 6] with in situ images from NASA’s Perseverance rover. We then developed Python-based algorithms in a GIS environment which considered factors such as topography (derived from HiRISE DEMs), geomorphology (from NOAH-H), and solar radiation balances at a test site within the nominal landing area. These data, and combinations thereof, were assigned weighting values that were passed to the algorithm and then used to compute individually optimized drive paths for different objective prioritizations.  

Using multivariate terrain analysis, our route-generation algorithms produced over thirty possible drive paths with associated statistics. The algorithm’s adjustable weighting parameters allow prioritization of variables, which will be critical when in situ data becomes available. We continue to iterate on our approach and will present current findings at this conference. Our work demonstrates that lightweight, flexible Python-based drive paths can be generated from existing data, supporting strategic planning and operational readiness across mission phases.

 

[1] Vago et al. (2017), Astrobiology, 17(6-7); [2] Volat et al. (2022), PSS, 222; [3] Fawdon et al. (2024), Journal of Maps, 20(1); [4] Bowen et al. (2022), PSS, 214; [5] Barrett et al. (2022), Icarus, 371; [6] Barrett et al. (2023),  Journal of Maps, 19(1); [7] Fawdon et al. (2021), Journal of Maps, 17(2); [8] Davis et al. (2023), EPSL, 601; [9] Favaro et al. (2021), JGR:P, 126(4); [10] Wright et al. (2022), Journal of Maps, 18(2).