| dc.description.abstract | Reliable in-field plant phenotyping requires a robotic platform capable of stable
terrain mobility, controlled motion, accurate navigation, synchronized sensing, and
consistent crop measurement under uneven ground, wheel slip, vibration, variable
loading, and dense vegetation. This research presents the ground-up design, fabrication,
system integration, and field operation of a four-wheel skid-steer agricultural robot
developed for blueberry-field navigation and three-dimensional plant phenotyping. The
complete platform was designed and built through this work, including the structural
frame, independent wheel-drive assemblies, adjustable suspension, power distribution
and electrical integration, embedded computing, communication, sensing, motion control,
navigation, and data-acquisition systems. The main technical contributions are an
adjustable terrain-responsive suspension, an integrated wheel-to-path control and
localization architecture, and a pose-regularized multi-camera reconstruction method for
plant-level height and canopy-volume estimation.
An adjustable air-spring suspension was designed, fabricated, and integrated into
the robot to support terrain compliance and sensing stability. Its dynamic behavior was
modeled and evaluated through a nonlinear quarter-car simulation incorporating
pressure-dependent air-spring stiffness, asymmetric damping, suspension travel limits,
and unilateral tire-ground contact. Under field-representative simulated terrain and
loading variations, the tuned suspension reduced root-mean-square chassis acceleration
by 11.04% and tire-unloading occurrence by 16.92% relative to the baseline configuration
while maintaining suspension displacement within the allowable range. A complete
motion-control and navigation architecture was developed by combining independent
wheel-speed regulation, actuator-current and torque constraints, skid-steer motion
conversion, path tracking, and multisensor localization using wheel odometry, inertial
measurements, and GPS. Hardware and field experiments demonstrated coordinated
four-wheel motion, controlled actuator demand, stable path tracking, and a 76.6%
reduction in localization error relative to raw GPS when evaluated against an independent
RTK reference.
The completed robot was deployed in blueberry fields to collect synchronized multi
camera data for three-dimensional canopy reconstruction. Repetitive foliage, severe
occlusion, limited camera overlap, and weak visual features made conventional image
based pose estimation unreliable. To address this problem, a pose-regularized
reconstruction framework was developed using calibrated camera parameters, fixed
inter-camera geometry, synchronized imaging, and sensor-derived motion information.
The constrained camera poses were integrated with 3D Gaussian Splatting to limit pose
drift and preserve canopy structure across multiple viewpoints. Ground-referenced plant
separation and geometric analysis were then applied to estimate individual blueberry
plant height and canopy volume. The method achieved normalized root-mean-square
errors of 4.77% for height and 6.68% for canopy volume, with coefficients of determination
of 0.883 and 0.960, respectively. The central contribution of this research is a complete
agricultural robotic system developed from platform design to quantitative crop
measurement, linking terrain-responsive suspension, four-wheel motion control,
multisensor navigation, synchronized field sensing, pose-constrained 3D reconstruction,
and non-destructive blueberry phenotyping within one operational framework. | en_US |
