Integrating 3D Sensing and Autonomous Robotics for Enhances Grassland Monitoring

Grasslands are challenging to characterize due to their homogeneous
visual texture and diverse species composition, complicating
automated data collection for precision livestock farming adoption.
Current methods lack detailed spatial information, hindering effective
grassland management. Recent advancements in industrial grade
field sensing technologies and 3D data processing provide new
opportunities for automated phenotyping.
We propose an integrated pipeline for grassland phenotyping using
an autonomous unmanned ground vehicle (UGV) powered by Robot
Operating System 2 (ROS2). The UGV follows predefined waypoints
using NAV2 to scan the grassland, capturing geo-referenced data. This
platform is designed to be easy to replicate, deploy and use, allowing
compatibility with a variety of sensors based on specific needs. It
features IP67-certified components, USB and Ethernet connectivity
for data transfer, and approximately 8 hours of battery life for active
operation.
Our current work focuses on combining RGB-D frames into a colorized
point cloud using a precise positioning device and registration
algorithms to assess sward height, LAI, biomass, and vegetation
indices while integrating species composition. Future work will
incorporate predictive modeling to assess climate and management
impacts on grassland health. Additionally, 3D rendering will simulate
virtual environments to optimize acquisition parameters before real
data collection. A semantic segmentation pipeline will be developed
using synthetic training data to spatially classify plant species and
objects of interest.
This approach integrates data acquisition, simulation, and predictive
modeling to enhance grassland monitoring and management,
enabling data-driven precision agriculture for sustainability and
productivity