Search

Agricultural robotics is a complex, challenging, and exciting research topic nowadays. However, orchard environments present harsh conditions for robotics operability, such as terrain irregularities, illumination, and inaccuracies in GPS signals. To overcome these challenges, reliable landmarks must be extracted from the environment. This study addresses the challenge of accurate, low-cost, and efficient landmark identification in orchards to enable robot row-following. First, deep learning, integrated with depth information, is used for real-time trunk detection and location. The in-house dataset used to train the models includes a total of 2453 manually annotated trunks. The results show that the trunk detection achieves an overall mAP of 81.6%, an inference time of 60 ms, and a location accuracy error of 9 mm at 2.8 m. Secondly, the environmental features obtained in the first step are fed into the DWA. The DWA performs reactive obstacle avoidance while attempting to reach the row-end destination. The final solution considers the limitations of the robot's kinematics and dynamics, enabling it to maintain the row path and avoid obstacles. Simulations and field tests demonstrated that even with a certain initial deviation, the robot could automatically adjust its position and drive through the rows in the real orchard. [ABSTRACT FROM AUTHOR]

Agricultural robotics is a complex, challenging, and exciting research topic nowadays. However, orchard environments present harsh conditions for robotics operability, such as terrain irregularities, illumination, and inaccuracies in GPS signals. To overcome these challenges, reliable landmarks must be extracted from the environment. This study addresses the challenge of accurate, low-cost, and efficient landmark identification in orchards to enable robot row-following. First, deep learning, integrated with depth information, is used for real-time trunk detection and location. The in-house dataset used to train the models includes a total of 2453 manually annotated trunks. The results show that the trunk detection achieves an overall mAP of 81.6%, an inference time of 60 ms, and a location accuracy error of 9 mm at 2.8 m. Secondly, the environmental features obtained in the first step are fed into the DWA. The DWA performs reactive obstacle avoidance while attempting to reach the row-end destination. The final solution considers the limitations of the robot’s kinematics and dynamics, enabling it to maintain the row path and avoid obstacles. Simulations and field tests demonstrated that even with a certain initial deviation, the robot could automatically adjust its position and drive through the rows in the real orchard.

Intra-row weeding is a time consuming and challenging task. Therefore, a rotary weeder implement for an autonomous electrical robot was developed. It can be used to remove the weeds of the intra-row area of orchards and vineyards. The hydraulic motor of the conventional tool was replaced by an electric motor and some mechanical parts were refabricated to reduce the overall weight. The side shift, the height and the tilt adjustment were performed by linear electric motors. For detecting the trunk positions, two different methods were evaluated: A conventional electromechanical sensor (feeler) and a sonar sensor. The robot performed autonomous row following based on two dimensional laser scanner data. The robot prototype was evaluated at a forward speed of 0.16 ms−1 and a working depth of 40 mm. The overall performance of the two different trunk detection methods was tested and evaluated for quality and power consumption. The results indicated that an automated intra-row weeding robot could be an alternative solution to actual machinery. The overall performance of the sonar was better than the adjusted feeler in the performed tests. The combination of autonomous navigation and weeding could increase the weeding quality and decrease power consumption in future.

In this paper, the effects of wheel slip estimation and compensation of trajectory tracking for orchard applications were investigated. A slippage estimator was developed and adapted into a car-like robot model. Steering and velocity commands were generated using a model-based control approach. The whole system was implemented and tested on an autonomous orchard vehicle that has steerable front wheels and actuated rear wheels. A high accuracy positioning system was used to estimate the longitudinal and lateral slip velocities while the vehicle is moving. A laser scanning range finder was placed at the front centre of the vehicle, which was used to detect rows of trees in the orchard. Procedures were first tested in a non-flat but open space, which was covered with snow. Then it was tested on an experimental orchard where the surface was covered with heavy mud and the vehicle was expected to follow trajectories that span multiple rows in the orchard. The vehicle detected individual trees as well as rows of trees to track the centre of each row and manoeuvred from one row to the next. The experimental results showed that trajectory tracking performance of the vehicle was enhanced via integrating a slippage estimator into the system model. Furthermore, using the slippage estimation in the system model increased the accuracy, repeatability and performance of the control system. [ABSTRACT FROM AUTHOR]

• Crop row following is especially challenging in narrow row cereal crops, such as wheat, where commercial row spacing may be as narrow as 125 mm. In cereal crops separation between plants within a row disappears at an early growth stage, and canopy closure between rows, when leaves from different rows start to occlude each other, occurs three to four months after the crop emerges. • Canopy closure makes it more challenging to identify separate rows through computer vision as clear lanes become obscured. Cereal crops are grass species and so their leaves have a predictable shape and orientation. • We introduce an image processing pipeline that exploits the grass growth form to identify cereal rows and track them. The key observation which is exploited is that leaf orientations tend to be more vertical along the rows and horizontal between rows due to the location of the stems within the rows. Adaptive mean-shift clustering on Hough line segments is then used to obtain lane centroids, and followed by a nearest neighbor data association method to gain the lane line candidates in 2D space. Lane parameters, e.g horizontal interception coordinates, are fit with linear regression and a Kalman filter is used for tracking lanes between frames. • The method is able to achieve less than 50 mm accuracy which is sufficient for placing a vehicles wheels between real-world, early-growth wheat crop rows to drive between rather than over them. Crop row following is especially challenging in narrow row cereal crops, such as wheat. Separation between plants within a row disappears at an early growth stage, and canopy closure between rows, when leaves from different rows start to occlude each other, occurs three to four months after the crop emerges. Canopy closure makes it challenging to identify separate rows through computer vision as clear lanes become obscured. Cereal crops are grass species and so their leaves have a predictable shape and orientation. We introduce an image processing pipeline which exploits grass shape to identify and track rows. The key observation exploited is that leaf orientations tend to be vertical along rows and horizontal between rows due to the location of the stems within the rows. Adaptive mean-shift clustering on Hough line segments is then used to obtain lane centroids, and followed by a nearest neighbor data association creating lane line candidates in 2D space. Lane parameters are fit with linear regression and a Kalman filter is used for tracking lanes between frames. The method is achieves sub-50 mm accuracy which is sufficient for placing a typical agri-robot's wheels between real-world, early-growth wheat crop rows to drive between them, as long as the crop is seeded in a wider spacing such as 180 mm row spacing for an 80 mm wheel width robot. [ABSTRACT FROM AUTHOR]

Autonomous navigation of field robots in an agricultural environment is a difficult task due to the inherent uncertainty in the environment. The drawback of existing systems is the lack of robustness to these uncertainties. In this study we propose a vision-based navigation method to address these problems. The focus is on navigation through a maize field in an outdoor environment where the robot has to navigate through a corridor formed by two plant rows, detect the end of the rows, navigate the headland and turn into another corridor under natural conditions. The method is based on a Particle Filter (PF) using a novel measurement model, where we construct a model image from the particle and compare it directly with the measurement image after elementary processing, such as down-sampling, excessive-green filtering and thresholding. The new measurement model does not extract features from the image and thus does not suffer from errors associated with the feature extraction process. We show how PF can be used for robust navigation of a robot in a semi-structured agricultural environment such as maize fields with inherent uncertainty. We demonstrate the robustness of the algorithm through experiments in several maize fields with different row patterns, varying plant sizes and diverse lighting conditions. To date we have logged over 5 km of successful test runs in which the robot navigates through the corridor without touching the plant stems, accurately detects the end of the rows and traverses the headland. [Copyright &y& Elsevier]

The purpose of this study was to develop a hydraulic control system (HCS) as a control unit around which an automatic furrow guidance system (AFGS) for tractor-mounted implements might be developed. The HCS consisted of a hydraulic cylinder slide device, a hydraulic pump, and an electrohydraulic proportional flow and direction control valve (solenoid valve). The basic specifications for the HCS are a hydraulic pressure of 6.8 MPa and flow capacity of 22 L min-1 for the pump, and the fastest speed of the double-acting cylinder, which is mounted to the lower link of a tractor's three-point hitch, needs to be 150 mm s-1. In order to evaluate the basic pressure control characteristic of the HCS, system responses were measured for a step and a simulated crop row input in the laboratory. In field tests, the implement was attached to the three-point hitch linkage of a tractor, with the tractor traveling speed set at 0.5, 1.0, and 1.5 m s-1, to evaluate the crop row and line marker tracking accuracy of the HCS. In the laboratory experiments, the computer needed 0.22 s to calculate and send the control signal to the solenoid valve, which then initiated the hydraulic cylinder movement in 0.07 s. For the frequency response, the cutoff frequency calculated was 0.25 Hz, and the phase delay was 56°. In other words, this guidance system was able to adjust the implements in 3.8 cm s-1. According to earlier research, the maximum adjustment speed permissible for crop row variation was 3.3 cm s-1. The RMS error of the HCS was less than 2.0 cm in tracking the line marker and less than 2.5 cm for crop rows in the field experiments. However, the implement tines did not actually engage the soil in the field experiments. [ABSTRACT FROM AUTHOR]

Intra-row weeding is a time consuming and challenging task. Therefore, a rotary weeder implement for an autonomous electrical robot was developed. It can be used to remove the weeds of the intra-row area of orchards and vineyards. The hydraulic motor of the conventional tool was replaced by an electric motor and some mechanical parts were refabricated to reduce the overall weight. The side shift, the height and the tilt adjustment were performed by linear electric motors. For detecting the trunk positions, two different methods were evaluated: A conventional electromechanical sensor (feeler) and a sonar sensor. The robot performed autonomous row following based on two dimensional laser scanner data. The robot prototype was evaluated at a forward speed of 0.16 ms−1 and a working depth of 40 mm. The overall performance of the two different trunk detection methods was tested and evaluated for quality and power consumption. The results indicated that an automated intra-row weeding robot could be an alternative solution to actual machinery. The overall performance of the sonar was better than the adjusted feeler in the performed tests. The combination of autonomous navigation and weeding could increase the weeding quality and decrease power consumption in future. [ABSTRACT FROM AUTHOR]