High-precision “autofarming”, or precise agricultural vehicle guidance, is rapidly becoming a reality thanks to increasing computing power and carrier-phase differential GPS (“CPDGPS”) position and attitude sensors. Realistic farm trajectories will include not only rows but also arcs created by smoothly joining rows or path-planning algorithms, spirals for farming center-pivot irrigated fields, and curved trajectories dictated by nonlinear field boundaries. In addition, fields are often sloped, and accurate control may be required either on linear trajectories or on curved contours.
A three-dimensional vehicle model which adapts to changing vehicle and ground conditions was created, and a low-order model for controller synthesis was extracted based on nominal conditions. The model was extended to include a towed implement. Experimentation showed that an extended Kalman filter could identify the vehicle's state in real-time. An approximation was derived for the additional positional uncertainty introduced by the noisy “lever-arm correction” necessary to translate the GPS position measurement at the roof antenna to the vehicle's control point on the ground; this approximation was then used to support the assertion that attitude measurement accuracy was as important to control point position measurement as the original position measurement accuracy at the GPS antenna.
The low-order vehicle control model was transformed to polar coordinates for control on arcs and spirals. Experimental data showed that the tractor's control, point tracked an arc to within a −0.3 cm mean and a 3.4 cm standard deviation and a spiral to within a −0.2 cm mean and a 5.3 cm standard deviation.
Cubic splines were used to describe curve trajectories, and a general expression for the time-rate-of-change of curve-related parameters was derived. Four vehicle control algorithms were derived for curve tracking: linear local-error control based on linearizing the vehicle about the curve's radius of curvature, linear finite-preview control using discrete linear quadratic tracking, nonlinear local error control based on feedback linearization, and nonlinear finite-preview control using nonlinear optimization techniques. The first three algorithms experimentally demonstrated mean tracking errors between zero and four centimeters and standard deviations of roughly four to ten centimeters. The fourth algorithm was computationally too expensive to implement with current technology. In experiment, the feedback linearization algorithm outperformed the other two control algorithms and also used the most control effort.
For control on sloped terrain, a variation on bias estimation (termed slope-adjusted bias estimation) was created, based on the terrain slope information calculated from vehicle attitude measurements. Slope-adjusted bias estimation demonstrated a 25% improvement in the standard deviation of the tractor's row-tracking error over “normal” bias estimation on terrain sloped at grades up to 28%. The CPDGPS attitude information was also used to develop a contour-tracking controller that tracked a contour to within a mean height error of 0.5 cm and a standard deviation of 4.3 cm without any prior knowledge of the terrain.
These real-time vehicle control results, applicable to any front-wheel-steered vehicle, demonstrate that accurate real-time control is possible over a variety of trajectories needed in a commercial autofarming system. This research is a significant step towards completely automating tractor control because farmers can now build global trajectories composed of the different types of trajectory “building blocks” developed here. Experimental results demonstrate that farmers can expect precision tracking down to the limit of the GPS position and attitude sensors.
[1]
John Dinwoodie,et al.
SAE Technical Paper Series
,
1987
.
[2]
Masayoshi Tomizuka.
Feedforward digital tracking controllers for motion control applications
,
1992,
Adv. Robotics.
[3]
G. M. Owen.
A Tractor Handling Study
,
1982
.
[4]
Weiping Li,et al.
Applied Nonlinear Control
,
1991
.
[5]
A. Laub.
A schur method for solving algebraic Riccati equations
,
1978,
1978 IEEE Conference on Decision and Control including the 17th Symposium on Adaptive Processes.
[6]
R. Dreisbach,et al.
STANFORD UNIVERSITY.
,
1914,
Science.
[7]
Helmuth Späth,et al.
One-dimensional spline interpolation algorithms
,
1995
.
[8]
J. Taylor.
An Introduction to Error Analysis
,
1982
.
[9]
R Buick,et al.
Comparing GPS Guidance With Foam Marker Guidance
,
1999
.
[10]
Drainage Division,et al.
Journal of irrigation and drainage engineering
,
1983
.
[11]
N E Manos,et al.
Stochastic Models
,
1960,
Encyclopedia of Social Network Analysis and Mining. 2nd Ed..
[12]
Aníbal Ollero,et al.
Mobile robot path planning for fine-grained and smooth path spcification
,
1995,
J. Field Robotics.
[13]
Vasile Sima,et al.
Algorithms for Linear-Quadratic Optimization
,
2021
.
[14]
E. Kreyszig,et al.
Advanced Engineering Mathematics.
,
1974
.
[15]
Jean-Claude Latombe,et al.
Robot motion planning
,
1970,
The Kluwer international series in engineering and computer science.
[16]
J. Speyer,et al.
On-line aircraft state and stability derivative estimation using themodified-gain extended Kalman filter
,
1987
.
[17]
Robert F. Stengel,et al.
Optimal Control and Estimation
,
1994
.
[18]
Jo Yung Wong,et al.
Theory of ground vehicles
,
1978
.