Progress 07/01/24 to 06/30/25

Outputs
Target Audience:Two graduate students were hired starting spring 2025 to work on the project. One graduate student (GRA 1) who was hired to conduct soil sensor development, gained skills in soil sampling techniques, including operating Giddings machine for core extraction, in identifying soil depths, and handling soil samples throughout processing. In addition, she increased her proficiency in SolidWorks for designing soil sensing components, supporting future sensor integration, gained hands on experience with different types of spectrometers, learning calibration and scanning techniques, and data analysis. She improved her skills in machine learning techniques such as PLSR, SVR, Random Forest, and Neural Networks for soil property prediction modeling and advanced my data analysis and spectral preprocessing abilities using Python. The second student (GRA 2) was hired to work on the Uncrewed Ground Vehicle (UGV) setup and integration. It was based on Robot Operating System (ROS) 2, and expertise in ROS2 was enhanced as a result. In addition, UGV navigation is a combination of different algorithms such as sensor calibration, Simultaneous Localization and Mapping (SLAM), path planning, simulation testing, hardware testing, and User Interface (UI) prompts. Experience in these individual sections was greatly improved and debugging and frequent testing of different setups helped to enhance overall understanding. All these sections were followed by a comprehensive literature review and productive discussions, which also contributed to personal growth during the project. Changes/Problems:The major challenge was that the project start was delayed by few months since the GRAs were hired in Spring 2025. So, soil sampling was prioritized during the summer of 2025 to collect the samples needed. Still, obtaining permission for sampling sites in private lands is still challenging. For that, assistance from local NRCS officials is requested. Since the delivery of the UGV was delayed, the system has been tested only in simulation so far. Typically, UGV simulation files are released by the manufacturer, but files for the Husky A300 were unavailable until May 2025. To address this, custom configuration files replicating the Husky A300's functionality were developed from scratch. These had some limitations due to the simulation environment, but efforts were made to match real-world specifications as closely as possible. What opportunities for training and professional development has the project provided?This project has provided GRA 1 with multiple training opportunities, including hands-on experience operating the Giddings machine, performing spectrometer calibration and scanning, and developing designs in SolidWorks, including 3D printing and mechanical component assembly. She also attended weekly and monthly research meetings with the PIs, where she presented her progress to the team, and participated in AI in Agriculture and Natural Resources conference (March 31 - April 2, 2025, Mississippi State University) and modeling technique webinars, which have strengthened my technical expertise and research communication skills. GRA 2 also attended the weekly and monthly research meetings with the PIs to present his work to the team and get feedback. In addition, he participated in the 2025 SPIE Defense + Security, held in Orlando, FL, offered exposure to industry innovations and peer discussions, especially around SLAM and autonomous navigation. Participating in technical sessions, post-session discussions and Q&A sessions, further enhanced the GRA's knowledge in this field. Furthermore, the GRA visited the exhibition stalls from various major companies and understood the importance of appropriate hardware and software implementation strategies. How have the results been disseminated to communities of interest? Nothing Reported What do you plan to do during the next reporting period to accomplish the goals?Soil sample collection will continue to collect additional samples adding up to ~300 processed samples. Models will be developed to estimate soil health properties from spectral data of the collected samples. Preliminary designs for soil sensors will also be developed. When the Husky A300 is delivered, systems currently tested in the simulation will be physically tested. This will help identify differences between simulation and real-world performance, allowing updates to the code and offering insights into simulation limitations and overlooked differences. The current navigation maps the terrain before moving to waypoints, but the goal is to navigate unknown terrain while collecting soil samples. This requires an exploratory approach using visual, LiDAR, and high-accuracy GPS data. Related work is expected to begin. The current path planning algorithm will be further tested and refined based on alternative strategies from the literature. Improvements to autonomous navigation, including using GPS (currently limited in simulation) as the main positioning method, are also planned.

Impacts
What was accomplished under these goals? Under Objective 1, six soil cores (up to 1 m depth) from two different locations in Mississippi were collected using the Giddings probe. Upon transportation to the lab, each soil core was segmented to 5 depths followed by separating a sub sample for laboratory chemical analysis. The remaining soil samples were processed (air dried and ground) and scanned using MIR ATR, diffuse reflectance, and other spectrometers under various conditions (fresh, 2 mm sieved, and fine ground). Soil sampling campaign is targeting 25 soil series in Mississippi and underway. For Objective 2, a Zirconia ATR probe was purchased and used for initial signal testing. An unexpected moisture absorbance was observed and is currently under investigation to improve signal quality. Since the UGV was not delivered during the reporting period, all setup was done in the simulation environment using the Linux version compatible with Husky A300 (Ubuntu 24.04) and compatible ROS packages. Configuration files were developed to replicate onboard sensor parameters within the limitations of the simulation platform. Two form factors, namely a scaled-down version of Husky A300 and a full-scale model, were implemented, which allows both current and future debugging and testing before deployment to physical hardware. Autonomous navigation was implemented using an open-source ROS2 package, and this functionality currently works only for pre-mapped terrains. A path planning algorithm was implemented as a standalone ROS2 package and tested in multiple versions to assess performance and suitability. A basic UI was developed to test ROS2 integration and platform compatibility. Initial controls included robot spawning, feedback monitoring, log displaying, and a master kill switch. This early design supported planning for the final version, focusing on usability and feature expansion.

Publications