Pesticides Research, 97 – Autonomous Weeder for Christmas Trees – Basic Development and Tests

Autonomous Weeder for Christmas Trees – Basic Development and Tests

Summary

Robotic weeding of Christmas trees is now a realistic possibility

Robotic weeding of Christmas trees is now a realistic possibility

Development work has shown that it is possible to develop a small driverless machine that can effectively control weeds in Christmas tree plantations.

Background and purpose

Control of weeds in Christmas tree plantations is essential for good growth and quality of the trees. Weeds are usually controlled by frequent spraying, which may have undesired environmental effects. Recent mechanical weed control methods have been developed, but these must be carried out with special machines, which are expensive to use, especially because of the high labour requirement.

The purpose of the present project was to develop and test a small, driverless weeding machine in order to alleviate these problems and make Christmas tree production more environmentally and economically sustainable. The starting point for development of the machinery concept was field experiments, which showed that it with regards to growth was sufficient to control weeds within a circle of 40 cm radius around each tree. This corresponds to about 40 % of the total area. Plants growing on the rest of the area may be left as they provide some shelter and may have a positive effect on the trees.

The investigation

The ACW (Autonomous Christmas tree Weeder) is based on a platform from a self-propelled 4-wheel grass mower of a size and shape suitable to operate between rows of Christmas trees and cut weeds around the trees. The machine is modified to include computer control of all major functions and the cutter was changed to suit weed cutting the tree canopy.

The modifications included replacement of the original centre mounted rotor cutter with a smaller unit mounted on a swinging arm, which could be moved into the row and under the branches to cut close to the trunks of the trees. The normal driver interfaces were replaced with linear actuators for operation of the clutch, brake, gear, steering, throttle, cutter drive clutch and cutter arm. The primary position sensor was a Real-Time Kinematic Global Positioning System (RTK GPS). These peripherals were connected to an embedded computer, which together with the other electronics were encapsulated in a weatherproof box on the front of the ACW. The computer communicated with the coordinator through a radio Ethernet connection (WiFi). Finally, side shields were fitted to the ACW to protect the trees from damage as the machine passed down the row.

This phase of the ACW project developed a deterministic control of the machine which was based on a precise map of the tree positions and the plantation boundaries. From this map a suitable vehicle route plan was defined along the tree rows. This route plan was used to control the machine movement in the rows, while the tree map was used to control the swing-arm rotor cutter. The plantation borders were used to make sure that the machine would only work within the plantation boundary.

The swing-arm cutter unit had a passive and an active control mode. In the passive mode the rotor was pulled out into the intra-row area by a spring. On contact with the tree trunk, the cutter unit would be pushed back, slide around the tree trunk and be released into the intra-row area again. The trees were shielded by a padded cover disc which had a slightly larger diameter than the rotating knives. In the active mode the cutter arm would pull in the rotor cutter when it approached a tree and releasing it back into the intra-row area when the tree had passed. This active mode was considered necessary for weeding in plantations with small, newly planted trees, which may be susceptible to being touched by the machine.

The control system was developed in the programming and simulation tool: Simulink from MathWorks. This tool was used to build integrated models of the vehicle as well as the control system. These models were used to simulate the functionality of the different elements and the complete system until it worked satisfactorily. When this was achieved the vehicle model was replaced by the real vehicle and tested physically with the control algorithm, first in a full scale model of a plantation and then to a limited extent in a real Christmas tree plantation.

The main conclusions

This work has shown that it is technically feasible to build an autonomous vehicle for weeding in Christmas tree plantations and similar environments. The control system was able to navigate the vehicle within a few cm from the planned track, based on RTK-GPS readings. However, in uneven terrain corrections from a tilt-meter were needed as undulating ground had a serious effect on the roll of the vehicle and therefore changed its indicated position.

In the best cases, weeds were cut right up to the trees and there were no significant damages, although repeated touching may cause problems.

Although the ACW functioned well, the reliability was not acceptable. Likewise, as the machine operation was totally deterministic no reactive behaviours were programmed into it. Hence the safety with respect to the surroundings was not satisfactory. However, the allied work has led to feasible solutions to these shortcomings and we now know what is needed to make this machine truly autonomous.

This concept seems promising for commercialization, since this kind of machine can also be used in other row crops. Thus it seems this method can contribute significantly to reductions in herbicide usage, as well as providing an improved work environment.

An economic estimate indicates that an autonomous weeder with the present price levels may be competitive with conventional mechanical weed control (about 3000 DKK/ha). The estimate also indicate that price reductions on electronical components and sensors assumed during the next 5 years may reduce the costs to a level competitive to spraying with herbicides (about 1500 DKK/ha).

Project results

In the model plantation, with an almost level surface, the ACW followed the route plan with cm-level precision. In passive mode the rotor cutter blades weeded very close to the tree trunks but less precise when the active control mode was used, although this could be improved with a better control model.

In the Christmas tree plantation, where the surface was uneven, the ACW had sometimes problems keeping to the route plan. This was due to the GPS antenna translating laterally as the machine rolled. However, the quality of the weeding by the cutter unit was excellent, when used in the passive mode. Almost all weeds around the trees were cut without damage to the trees. To achieve acceptable machine precision under rough conditions it is necessary to use a tilt-meter in combination with the GPS.

The trials also showed that the reliability and safety of the ACW needs significant improvements to be acceptable for unsupervised work. Quite often the vehicle stopped (as it should) due to insufficient GPS quality (usually due to radio shadow from large trees or simply due to lack of enough visible satellites). To avoid this problem it is necessary to supplement the GPS (absolute positioning) with at least one relative positioning system such as an Inertial Navigation System (INS) or a local proximity sensor such as a 2D laser scanner, which can detect the row position and orientation relative to the ACW. Furthermore it is necessary to have “sensor fusion” of the signals from this extra sensor and the GPS in a Kalman filter. Finally, it is necessary to have a sensor which can detect large obstructions in front of the ACW, and a bumper with simple force sensors for collision detection.