MTRI INC — Department of Defense STTR Phase II: A21C-T021

Description

NATO is currently developing the Next-Generation Reference Mobility Model (NG-NRMM), which is used to determine terrain strength properties for mobility. The NG-NRMM relies heavily on in-situ measurements, which is a significant limitation for traversing unknown areas and combat zones. The MTRI Inc. and MTU team recently competed the A21C-T021 Phase I Small Business Technology Transfer (STTR) project, where the team demonstrated that unmanned aerial vehicle (UAV) remote sensing data can be used to determine soil strength properties. The focus of Phase I was to predict the cone penetration value for the 0-6 inches (CP06) and the 0-12 inches (CP12) ranges. Through Phase I the project team demonstrated a proof-of-concept system that can that can map terrain strength properties and soil moisture from a drone. This system utilized UAV-based data from hyperspectral, multispectral, and thermal sensors to determine soil characteristics and apply machine learning and deep learning algorithms to predict CP06 and CP12. A final two-day demonstration focused on a near real time offline method that produced terrain strength estimates within 24 hours. The remote sensing data collected on the first day was processed and analyzed to produce maps of terrain strength and soil moisture before the morning of the second day. On the second day, an MRZR was used to demonstrate the accuracy of the predicted CP06 and CP12 maps. This Phase II STTR will focus on turning the offline methods developed in Phase I into a combination of UAV-based offline assessment and vehicle-based online assessment for real-time autonomous vehicle mobility. The offline assessment will be performed using UAV-based hyperspectral and thermal cameras to determine the soil properties for terrain strength models on the same day they are used. This data, along with the trained reference models, will be uploaded to a computer onboard an autonomous vehicle (i.e., MRZR). Then, during the online portion, the trained reference models will be used in conjunction with real-time data from a thermal camera for vehicle routing. To accomplish this, a robust machine learning or deep learning model needs to be developed which can be applied to a wide variety of terrain, soil types, and soil moisture conditions. The project team will be performing extensive field data collections at multiple sites. This will not only aid in improving the accuracy of the models developed in Phase I, but also allow for greater portability to uncharacterized locations. The project team will develop the metrics and methods to be used to evaluate the benefits of the real-time remote estimation of terrain strength. Multiple scenarios will be conducted in a variety of soil types and conditions. The successful completion of this project will demonstrate how the real-time remote estimation can improve an autonomous vehicles ability to navigate through terrain with varying soil strength properties and in a more efficient manner than without.