Friday, May 17, 2013

design options, and factors which impinge upon them

This is the second post in a series, which began with what operations should a ‘cultibot’ be able to perform?.

If you've read the masthead for this blog, you'll know that it is about how robotics can assist with the application of horticultural methods on an agricultural scale, so, even though I've been talking about gardening, the real goal is to apply gardening methods (or at least the attentive attitude of a careful gardener) over the entire landscape, ‘from sea to shining sea’, wherever land is managed as opposed to being left entirely wild. That's a tall order, of course, and no single system could conceivably cope with the vast range of conditions implied.

So, to simplify matters somewhat, I'll confine myself to consideration of land primarily engaged in production for market, whether for food, fiber, fuel, or any of the other reasons we put land to productive use, and to systems engaged in the management of land for production.

In this context, it's reasonable to divide plants into two basic categories, crops and weeds. Any such system must be able to differentiate between the two and nurture crop plants while eliminating or at least suppressing the growth of weeds. Differentiating plants is more easily accomplished after they have grown beyond the seedling stage, but controlling weeds that grow from seed is more easily done while they are still seedlings. So one design choice is whether to attempt to differentiate seedlings and eliminate weed seedlings before they've had a chance to root deeply.

Differentiating seedlings can be made considerably easier if you know very precisely where the crop seeds have been planted, so anything appearing more than a very short distance from one of these locations has a high probability of being a weed. Beyond that, a robotic gardener could use visual information (from gross morphology to the vein pattern of the seed leaves) and/or sniffing for volatile compounds. Visual information can be acquired from a considerable distance, relative to the size of the seedling, whereas chemical sniffing requires close approach, and is generally better accomplished by a small device.

Either way, once the decision to take out a seedling has been made, there remains a choice of methods. If the device which detected it is small, light, and locally supported, meaning that its weight rests on nearby soil, then a mechanical extraction of the seedling from the soil (uprooting it) may be the most reasonable approach, the main concern being that that it grasps the correct seedling.

On the other hand, if the device is remotely supported, meaning that it is essentially suspended amid the foliage, mechanical uprooting may still be an option, if it also possesses one or more quick-moving appendages, but the time involved will be a more pressing consideration, as a remotely supported device would necessarily represent a larger investment, so there will be fewer of them and each must handle a larger area. But other options are available to a remotely supported device. It might use a laser which can be retargeted as fast as a mirror can be repositioned, or a miniature turret that fires ice pellets, and if it uses an arm, that arm might slice through the seedling with a thin jet of high pressure water rather than grasping it.

So much for weed seedlings, but they're the easy example. Let's consider weeds, like some grasses and shrubs, that spread by rhizome. A rhizome is a horizontal root, running just below the surface, occasionally sending up stems which appear from above to be independent plants, but which are actually part of a larger, robust organism. If a plot is heavily infested with plants that propagate by rhizome, they cannot be eliminated without also destroying whatever crop was in the ground at the time. The approach that works most reliably is to frequently survey for emerging stems of these plants, pulling them up with as much of the rhizome as comes along fairly easily, and to keep at it until the rhizomes run out of stored energy for putting up new stems and rot in the ground. This is an operation that really can't reasonably be performed by a locally supported device, because it can require significant force to pull up such a stem, much less the attached rhizome. A locally supported device might clip them off at the soil surface, but to be effective, that would need to be repeated even more frequently than the uprooting.

Something a small, locally supported device might do would be to apply very small doses of some specific herbicide, such that it would be absorbed into the rhizome itself before killing the stem, but this begs the question of whether herbicides are to be used at all. (If they are to be used, then direct application of small doses to the weeds to be controlled is the obviously best approach.) Something a remotely supported device could do that a locally supported device really couldn't, would be to use an electromagnetic sensor to trace a signal transmitted from one section of rhizome to locate another spot where it could be exposed and then heated by passing an electrical current through it. Another option would be to inject steam next to the stem, killing the stem and a short segment of rhizome.

If we were to go on to consider other operations, for example harvest and dealing with the coarse plant material left over after harvest, it would quickly become apparent that the larger, remotely supported machines are necessary, whereas the smaller, locally supported devices are probably optional, although the most efficient approach might well be to use both in a complementary fashion, the gardening system mentioned earlier, perhaps even including insect-sized devices that fly among or crawl over the plants themselves.

Strictly speaking, devices suspended from a wheeled platform that always follows the same tracks, a practice which is beginning to gain traction in precision agriculture circles, would qualify as remotely supported for the purpose of this discussion, even though part of the surface of the field is sacrificed to supporting the weight of the machines. Where the wheels of those machines don't run, the soil remains uncompressed. Placing gravel on the tracks they do use would help limit damage. Another option would be to install rails or troughs that double as channels for water transport. Still another option would be to support the platform on legs that only ever placed their weight on particular spots. Any of these options would help avoid the rutting of continuous paths that could lead to erosion.

Other design choices include sensors and the sorts of information gathered and the types of manipulators provided. A garden is a very complex environment, most of which is delicate, and while that delicacy makes the design and control of manipulators more difficult than it would otherwise be, getting adequate information into the machine, and deciding what operations to perform, is by far the greater challenge.

Laser scanners of various sorts could prove very useful in mapping foliage. Imaging radar could help locate stems that are hidden by foliage. Video, particularly if high definition not only in pixel density but in color depth (and ideally in splitting the spectrum more finely than the usual RGB), combined with telescopic optics, could make differentiation of seedlings far easier. Chemical sensing, mentioned above, could be used, for instance, to determine the ripeness of fruit and to detect the presence of some infestations and diseases. Very acute hearing could help with the discovery of insect activity. Infrared video would help with locating warm-blooded animals. Touch would be important in making it possible for manipulators to reach through foliage or pick fruit without damage.

All that said, the most fundamental choice isn't any of the above, but rather what sort of management regime to support. I'll address this issue in the next installment.

No comments: