Around two years ago, we were contacted by Antoine Richard, a researcher from the University of Luxembourg working on a paper relating to ML vision-based navigation of lunar rovers. As there is no GPS on the moon, rovers require other methods of spatial positioning if they are to be automatically maneuvered. Antoine was interested in a machine-vision approach, which required a set of synthetic training data created from somewhat accurate lunar regolith (moon dirt) textures. He contacted us to propose a collaboration partially sponsored by the University and Spaceport Rostock, where we would fly to their lab in Rostock, Germany, to scan the regolith simulant. Once we were able to cover our own travel costs thanks to support from our Patreon supporters, we had free rein for a week to scan as much as we could in their facility. The Simulation Antoine himself accompanied us to assist in the layout of the material to simulate lunar conditions as much as possible, along with Frank Koch and Lasse Hansen from Spaceport Rostock. This involved careful scattering of the dust with various techniques, using mostly gardening and kitchen tools to attempt to recreate the nature of the moon’s surface. Although the simulated regolith is made in such a way as to copy the substance on the moon as closely as possible – by grinding specific types of rock into extremely fine powder – the moon has no atmosphere, and so the dust behaves rather differently than it does here on earth. Because there is no friction to slow the dust down once it becomes airborne, for example after a meteor impact, the surface is riddled with tiny craters. The Danger Likewise due to the lack of atmosphere, the regolith has little to abrade the sharp edges of the dust particles away. This makes the fine power extremely rough and easily scratches any sensitive camera equipment that it comes into contact with, even in the air. It also makes it carcinogenic, so breathing the dust is dangerous to your health. Dealing with this threat was one of the primary challenges of the trip, as we had to wait for the dust to settle (literally) after each arrangement, which took 1-2 hours, and then be careful not to disturb the surface while we scanned it. We were also required to wear N95 face masks while working near the material, and employed several air quality monitors which notified us when we were being too rigorous with the arrangement of the dust. The Process For normal terrestrial scans, it is typical to walk back and forth across the surface you are scanning to capture a dense grid of images. These images are then fed into a photogrammetry software for mesh reconstruction. But as we could not physically walk across the regolith without disturbing dangerous clouds of dust, we constructed a single axis automated gantry which could suspend our camera and flash and automatically scan a row of images. Then from the sides, we could manually move the rig along the other axis. Each 7×4 meter scan took approximately 2-3 hours to capture this way. On the side, we also captured smaller 40cm macro scans of the regolith, intended to be used for smaller-scale renders or in combination with the larger scans to add detail. The Data Each scan consisted of approximately 1500 photos, totaling 1 million megapixels (or a terapixel if you like) and resulted in reconstructed geometry made of 1.7 billion polygons. A screenshot of one of the surface scans. By the end of the week, we had successfully scanned 20 textures. As part of our promise to Antoine, we are also releasing all the raw data under the public domain (CC0) for any other researchers who wish to take advantage of it, or any photogrammetry artists who want to build their own geometry from the scans. The total dataset is ~800GB, but you can download each scan individually: moon_01 + moon_02 (79 GB) moon_03 + moon_04 (87 GB) moon_dusted_01 (34 GB) moon_dusted_02 (48 GB) moon_dusted_03 (41 GB) moon_dusted_04 (38 GB) moon_dusted_05 (34 GB) moon_flat_macro_01 (35 GB) moon_flat_macro_02 (8.1 GB) moon_footprints_01 + moon_footprints_02 (49 GB) moon_macro_01 (38 GB) moon_meteor_01 + moon_meteor_02 (91 GB) moon_rock_01 (5.1 GB) moon_rock_02 (4.4 GB) moon_rock_03 (4.9 GB) moon_rock_04 (4.8 GB) moon_rock_05 (4.6 GB) moon_rock_06 (4.6 GB) moon_rock_07 (4.5 GB) moon_tracks_01 + moon_tracks_02 (40 GB) moon_tracks_03 + moon_tracks_04 (37 GB) Please do let us know if you do anything interesting with this data, we and the Spaceport Rostock team would love to hear about it. If you have any trouble downloading the files, we also have them on a Nextcloud server that we can share with you. Just get in touch and we’ll give you access freely. Additionally, we scanned 7 regular boring terrestrial rocks coated in the regolith to make them appear more like lunar rocks in order to help visual reconstruction of a simulated lunar environment. For fun, we captured an HDRI in the middle of the regolith pit – a precarious adventure – and lit it using their monstrous halogen bulb meant to simulate the high-contrast low-angle lighting often found on the moon. With the light reflected off the white ceiling you don’t quite get the same effect, but it was interesting to capture nonetheless. The Assets Here is the full collection of assets we created in this project: The Demo As with all of Poly Haven’s other asset collections, we like to showcase what artists can do with our assets by creating a render. James took all of the scans that the team has put together over the last two years and built a beautiful simulation of a lunar environment in Blender. The scene file for this render will be made available soon.