Autonomous Agricultural Vehicles Research Articles

Image Processing for Ridge/Furrow Discrimination for Autonomous Agricultural Vehicles Navigation

Autonomous navigation is important to robotize agricultural vehicles. This paper discusses a development of an image processing method to discriminate between traversable region (furrow) and non-traversable region (ridge) in ridged fields with no crops. Our method consists of 3 parts. First, images were separated into images with shade and images with no shade by gray level histograms. Next, as for images with shade, colour information was used to discriminate between traversable region and non-traversable region. On the other hand, as for images with no shade, texture was focused on. In this research, variance was used as a texture description. The proposed method was tested on images acquired by a camera, taken of different fields and under different light conditions. Our method discriminated between traversable region and non-traversable region successfully.

A distributed control framework for motion coordination of teams of autonomous agricultural vehicles

A distributed control framework intended to coordinate the motions of teams of autonomous agricultural vehicles operating in proximity is presented; master–slave and peer-to-peer operation modes are supported. Each vehicle has a nonlinear model predictive tracking controller, which keeps it as close as possible to the path demanded by the task, and coordinates and avoids collisions with nearby vehicles. To do this, it receives the motion trajectories of all other vehicles in its vicinity that may interfere with its own projected motion via a wireless network, and incorporates these trajectories in the computation of its own optimal control action. Each controller is supervised by a higher-level task controller that determines a limited set of the controller's parameters. Simulation experiments have shown that the minimisation of the tracking error along a finite horizon enabled the controller to track paths containing sharp turns, by applying appropriate steering well in advance of the turn. It has also been shown that the variation of specific predictive controller parameters results in a wide range of behaviours; i.e., vehicles can move in both operating modes, coordinate with nearby vehicles by altering their velocity profiles or the shapes of their paths, and avoid collisions.

Double look-ahead reference point control for autonomous agricultural vehicles

A simple automatic path tracking system is developed to satisfy the typical requirements of an unmanned agricultural tractor application based on two look-ahead reference points (2-LARP) on the desired path. The main objective of the proposed control system is to track a desired path within reasonable tolerances of a typical farming process including considerable slippage. The employed look-ahead reference points provide compensation for the centrifugal forces and reduction of the peak lateral deviation due to curvature transition, using only simple arithmetic operations. Extensive numerical tests were carried out on the computer simulation of the system dynamics driven by the proposed control method. Simulation results indicate enhancements in vehicle manoeuvrability. Tests indicated that the peak lateral displacement error at the curvature transitions is reduced to one fifth of single LARP error. ► Simple path tracking for agricultural vehicles using 2 look-ahead reference points. ► Lateral deviations similar to an exponentially decaying 6th order dynamic system. ► One look-ahead reference compensates for centrifugal forces over large speed range. ► Second look-ahead reference reduces the peak deviation at curvature transitions. ► The proposed controller reduces the peak error at curvature transitions.

Infield logistics planning for crop-harvesting operations

Crop-harvesting operations are typically carried out with combine harvesters. The harvested product is transferred to one or more tractors every time the combine harvester's storage capacity is reached. The efficiency of the process can be significantly improved by computing optimal routes and interactions for the harvest vehicles in the field. Furthermore, an automated method for generating itineraries for the harvest vehicles facilitates the planning for autonomous agricultural vehicles. The infield logistics problem is formulated as an integer linear programming vehicle routing problem with additional turn penalty constraints, but, because of the high number of decision variables, it is not possible to solve cases of realistic field size. The solution time of the infield logistics problem is considerably reduced by reformulating it as a modified minimum-cost network flow problem. This specific structure allows the exact solution of intermediate-size planning problems in a much shorter time period. The result of solving the infield logistics problem with the proposed modelling approaches is a set of itineraries (‘tours’), covering the entire field. Each ‘tour’ is characterized by the combine harvester's start and end points and the positions where the combine harvester needs to be unloaded. The planning models minimize non-productivity (i.e. the time when a combine harvester travels in a field without harvesting). The results indicate that coordination between combine harvesters and tractors is also improved.

A two-stage optimal motion planner for autonomous agricultural vehicles

This paper presents a two-stage motion planning algorithm which can compute low-cost motions for autonomous agricultural vehicles, for a given cost function defined over the entire path (e.g., shortest path, maximum clearance, etc.). In the first stage, the algorithm utilizes randomized motion planning to explore the space of possible motions and computes a feasible sub-optimal trajectory. In the second stage, the optimization of the stage-1 motion is formulated within the optimal control framework and function-space gradient descent is used to minimize the cost of the entire motion. The numerical results suggest that the two-stage motion planner can compute optimal or quasi-optimal motions in free space very quickly. In the presence of obstacles however, the execution time increases significantly. Furthermore, kino-dynamic, or dynamic motion models seem to be necessary in order to produce smooth motion trajectories.

CONTROLLER AREA NETWORK BASED DISTRIBUTED CONTROL FOR AUTONOMOUS VEHICLES

The goal of this project was to evaluate the potential of a controller area network (CAN bus) to be used as thecommunication network for a distributed control system on an autonomous agricultural vehicle. The prototype system utilizedmicrocontroller-driven nodes to act as control points along a CAN bus. Messages were transferred to the steering,transmission, and hitch control nodes via a task computer. The task computer utilized global positioning system data togenerate appropriate control commands. Laboratory and field testing demonstrated that each of the control nodes couldfunction simultaneously over the CAN bus. Results showed that the task computer adequately applied a feedback controlmodel to the system and achieved guidance accuracy levels well within the desired range. Testing also demonstrated thesystems ability to complete normal field operations, such as headland turning and implement control.

AN OBSTACLE IDENTIFICATION ALGORITHM FOR A LASER RANGE FINDER-BASED OBSTACLE DETECTOR

The capability of detecting and identifying obstacles on the expected path and taking appropriate collisionavoidance actions automatically is critical for safe operation of autonomous agricultural vehicles. This article presents anobstacle detection and identification algorithm for an on-tractor laser range finder-based obstacle detector. This algorithmconsists of a template matching function and a Kalman filter for detecting the location of an obstacle, reconstructing thesilhouette of the detected obstacle, and estimating its relative motions. Field validation test results verified that this obstacledetector was capable of detecting a moving object within a semicircle of 8 m radius and reconstructing a 2D silhouette ofthe obstacle progressively in real time. The errors of this obstacle detector in estimating the position, speed, and movingdirection of the obstacle relative to the tractor were 0.052 m, 0.11 m s-1, and 1.2, respectively. Such accuracies are sufficientfor providing safety warning and collision avoidance for autonomous tractors in field operation.

Heuristic optimization methods for motion planning of autonomous agricultural vehicles

In this paper, two heuristic optimization techniques are tested and compared in the application of motion planning for autonomous agricultural vehicles: Simulated Annealing and Genetic Algorithms. Several preliminary experimentations are performed for both algorithms, so that the best neighborhood definitions and algorithm parameters are found. Then, the two tuned algorithms are run extensively, but for no more than 2000 cost function evaluations, as run-time is the critical factor for this application. The comparison of the two algorithms showed that the Simulated Annealing algorithm achieves the better performance and outperforms the Genetic Algorithm. The final optimum found by the Simulated Annealing algorithm is considered to be satisfactory for the specific motion planning application.

Relating Torque and Slip in an Odometric Model for an Autonomous Agricultural Vehicle

This paper describes a method of considering the slip that is experienced by the wheels of an agricultural autonomous guided vehicle such that the accuracy of dead-reckoning navigation may be improved. Traction models for off-road locomotion are reviewed. Using experimental data from an agricultural AGV, a simplified form suitable for vehicle navigation is derived. This simplified model relates measurements of the torques applied to the wheels with wheel slip, and is used as the basis of an observation model for odometric sensor data in the vehicle's extended Kalman filter (EKF) navigation system. The slip model parameters are included as states in the vehicle EKF so that the vehicle may adapt to changing surface properties. Results using real field data and a simulation of the vehicle EKF show that positional accuracy can be increased by a slip-aware odometric model, and that when used as part of a multi-sensor navigation system, the consistency of the EKF state estimator is improved.

Heuristic Motion Planning for Autonomous Agricultural Vehicles

In this paper, two heuristic optimization techniques are tested and compared in the application of motion planning for autonomous agricultural vehicles: Simulated Annealing and Genetic Algorithms. Several preliminary experimentations are performed for both algorithms, so that the best neighborhood definitions and algorithm parameters are found. Then, the two tuned algorithms are run extensively, but for no more than 2000 cost function evaluations, as run-time is the critical factor for this application. The comparison of the two algorithms showed that the Simulated Annealing algorithm achieves the better performance and outperforms the Genetic Algorithm. The final optimum found by the Simulated Annealing algorithm is considered to be satisfactory for the specific motion planning application.

Ground based sensing systems for autonomous agricultural vehicles

This paper examines ground based (as opposed to satellite based) sensing methods for vehicle position fixing. Sensors are considered in various categories, motion measurement (odometry, inertial), artificial landmarks (laser positioning, millimetre wave radar), and local feature detection (sonar, machine vision). Particular emphasis is paid to technologies which have proven successful beyond the field of agriculture, and to machine vision because of its topicality. The importance of sensor fusion, using a sound theoretical framework, is emphasised. The most common technique, the Kalman filter, is outlined and practical points are discussed. As an example system, the autonomous vehicle developed at Silsoe Research Institute is described. This vehicle does not use an absolute positioning system, rather it navigates using local features, in this case the crop plants. This vehicle uses a sensor package that includes machine vision, odometers, accelerometers, and a compass, where sensor fusion is accomplished using an extended Kalman filter.