Multiple Rover Efficient Automated Mapping

Stottler Henke’s Multiple-Rover Efficient Automated Mapper (M-REAM) will provide the highest resolution and most frequently updated maps of the Lunar South Pole surface changes as a result of autonomous operations with minimal to zero impact on the primary science mission of those surface assets. This will bolster both the autonomous operations on the surface accuracy and science on the ground with the highest-quality digital twins of the Lunar South Pole operations. Our approach is divided into two components: an innovative 3D collaborative SLAM technique and scoring the “interestingness” of certain areas to adjust path planning algorithms. We will implement a custom-designed SLAM approach for the challenges of the Lunar South Pole from the low solar angle and long shadows. A learned Neural Radiance Field (NeRF) representation of the lunar regolith and localizations for rovers will be augmented by a Neural Visibility Field (NeRV), treating materials as reflectors of light rather than emitters. While we primarily update the lunar mapping utilizing images that a rover captures throughout its typical daily routine, we introduce a scoring mechanism for “interestingness” that allows for minimal adaptation of the rover’s primary mission to capture massive amounts of new information. For example, a RASSOR taking the same path from digging to the regolith processing site would be incentivized to take a slightly different route to update areas of the map that might be interesting. Interestingness is largely defined by autonomous operations in the area by a surface asset but is also driven by pixel-wise errors rendered by the NeRV NN, a learned time-decay function to account for meteorite impacts and other unexpected changes, or manually by a human operator. By developing, leveraging, and integrating these two cutting-edge technologies, the system will push the present boundaries of mapping on the lunar surface and enable more sophisticated science and missions.