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

Autonomous Weeder for Christmas Trees – Basic Development and Tests

8 Discussion

8.1 The ACW platform
8.2 The weed mechanical cutter concept
8.3 Control system
8.4 Tree mapping and cutter head control
8.5 Economic feasibility

The development work reported above has shown good and encouraging results but has also clearly indicated that more work is needed before the concept is reliable and safe enough for commercialisation. In the present chapter these issues are discussed in relation to the different elements of the vehicle.

8.1 The ACW platform

The platform selected for the ACW was found to be well suited for the purpose. The width fits well in between rows placed at the normal distance of 1.2 m, and the chassis ground clearance was found sufficient for operation in the test plantation, which was established on even farmland. However, it would have been an advantage to have a smaller turning radius to ease turnings. Also it would have been an advantage to have slightly larger wheels.

8.2 The weed mechanical cutter concept

The applied cutter concept (fig. 4.5 and 4.6) worked satisfactory in the rather limited tests performed so far. The only improvement that appears to be needed is the durability of the cutter knives. The knives used were made from a kind of hard plastic material but not strong enough to resist contact with stones. No tests were performed in areas with harvest residues from a previous tree crop, but it is unlikely that it would be able to handle it. If this is required a special designed and stronger cutter head would be needed. Alternatively, the residue could be removed or chopped before planting of new trees.

8.3 Control system

The tests have shown that improvements are needed both in the vehicle path tracking control and the cutter arm positioning control in order to have reliable operation.

8.3.1 Vehicle tracking control

As the vehicle control is important both for proper operation and for precise positioning of the cutter head it should be reliable under all reasonable conditions.

Our attempt to apply the existing tilt-meter was not successful because of its unsuitable dynamic characteristics. This problem has to be solved as one of the first things.

The other problem that the GPS-signal occasionally is of insufficient quality can be solved by addition of a supplementary sensor and fusion of the signals in a Kalman filter (see below). The simplest supplementary sensor would be wheel encoders, but these are only useful over short distances because of error accumulation.

8.3.1.1 Laser range scanner as row detector

Another possibility would be a laser range scanner as already mentioned in section 6.1.3. This sensor appear to be suitable for measurement of the distances to trees in the rows ahead, which can be transferred into vehicle orientation relative to the rows.

8.3.1.2 Kalman filter

To have optimum utilisation of sensor information and compensation for sensor faults and drop outs it is necessary to apply a Kalman filter, which on a statistical basis can merge the information from several sensors and get the best possible values for the control system (Welch and Bishop, 2003). Initial work has already been done on this in a student project, but further work is needed to adapt and integrate it into the ACW control system.

8.3.2 Weed cutter arm control

Also the cutter arm control should be more accurate. The current controller is of the “quasi-static” type, i.e. the mode is changed when the perceived distance to the nearest tree is below or above some threshold. A considerable improvement may be achieved on basis of previous simulations of the movements in response to certain actuator impulses before selecting the most appropriate impulse and time combinations for precise positioning.

8.3.3 Vehicle fault and obstruction detection and system reaction

In table 5.1, a list of vehicle fault and corresponding response was defined for the ACW. So far the following fault detection functions were made:

Actuator overload or actuator not working,
Gear setting error,
Engine not running or turning too slow,
GPS readings of insufficient quality,
ACW outside plantation,
Emergency stop button activated.

The system reaction to all of these is just: Stop of the engine. An important future task would be to build in the relevant reactions (Table 5.1) as well as further detection and reaction abilities as described in the following.

8.3.3.1 Detection of large obstructions

The ACW should be able to detect larger obstructions like people or animals in front of the machine. A possible candidate for this purpose would be the laser range scanner discussed above, which on basis of the initial investigations (fig. 6.1) appear to be suitable for detection of obstructions in the inter-row strip. It may also be suitable for detection of “negative obstructions”, i.e. deep tracks or water filled holes. Future investigations are needed to see if this is possible.

8.3.3.2 Detection of collisions

Even when much effort is used to prevent contact with obstructions, collisions may still occur. Therefore, it is necessary to have a mechanical bumper with switches or gauges that can detect forces above a certain threshold. If such an activation is detected the vehicle should stop immediately.

8.3.3.3 Detection of engine overload

The engine RPM sensor may be used for detection of vehicle overload. A sudden change of speed may indicate some sort of blockage, e.g. of the rotor cutter.

8.3.3.4 Detection of rotor cutter operation problems

The rotor cutter rotational speed should be measured continuously to detect problems like power transmission failure (e.g. v-belt broken or slipping), or broken cutter knife. A sudden change of speed will indicate a sudden overload, which for instance could be caused by some sort of blockage.

8.4 Tree mapping and cutter head control

The control of an ACW may be based on tree mapping at two different precision levels:

Highly precise mapping of all trees
Precise mapping of row end trees (or of trees at the ends of right lined sections of rows)

The first of these levels would allow active, touch free weed cutter head positioning close to each tree even if rows are curved or somewhat irregular. This mapping method would make it possible to do autonomous weeding during the complete growing period of trees.

The second level would be sufficient to define a proper route plan for the simpler passive cutter head control, i.e. the cutter head would hit each tree and slide passively around it providing good weeding but also a risk of damage to young trees. This control method would therefore only be appropriate if the trees are sufficient resistant, which may be the case 2 years after planting. Before this time weeding may be done with a special spring tooth harrow, which already is used like that in practical conditions.

Another alternative might be to control the ACW by means of a row detecting sensor, although this most likely only would be precise enough for passive cutter control.

8.4.1.1 Automatic mapping during planting

As the detailed manual mapping of trees is rather costly an automatic method would be of interest. One such alternative method would be to map tree positions during planting by recording of the positions (RTK-GPS) where the trees are placed in the ground. A system like this has already been developed for mapping of the placements of sugar beet seeds (Griepentrog et al. 2003).

8.4.1.2 Mapping by use of laser scanner

Another possibility would be automatic measurement of position of already planted trees using the ACW itself with a laser scanner and a mapping software system (Folkesson and Christensen, 2004). This would mean that the autonomous weeder (or another vehicle) would have to be operated manually through the plantation for mapping before the route planning and start of weeding can take place.

8.5 Economic feasibility

The result of the feasibility study (chapter 7) is of course very dependent on whether the assumed parameter values can be achieved. One of the most uncertain figures are the machine capacity and the running cost. For instance if service personel is far away, and the frequency of faults is high the machine capacity would be low and the repair costs high. On the other hand if a few more man years are spent on improving the reliability and safety, the estimates may well be realistic.