Wednesday, June 19, 2019

The ebb of ebb-and-flow

I was already trying to concentrate on my music-related programming project before WWDC (June 3rd through 7th), and now my enthusiasm for it has been renewed, so that's where I'll be putting my time and energy for a while, months at least.

That doesn't necessarily mean this blog will fall entirely quiet, but that is one possibility. If nothing meriting long-form treatment occurs to me, I won't be posting here for a while.

In the meantime, I do expect to post the occasional tweet related to the same theme as this blog. To see those, have a look at

Thursday, November 15, 2018

Logistics for distributed, diverse production

Farming operations don't exist in a vacuum. Farm products feed into a processing and distribution network, and, if nutrients are to be recycled, waste materials need to come back to farms to maintain and improve soils. Both flows, outward and inward, are served by keeping the supply lines (the need for transportation) as short as can reasonably be managed, given other constraints like economies of scale in processing, storage, distribution, and marketing.

This post will focus on the outward flow, from farm to processing and storage to distributor to vendor to customer, which, I'd argue, applies even to urban farms with direct-to-customer relationships, albeit in compressed form.

The usual model is that farms specialize, sometimes producing only a single commodity and typically only a few, with production profiles that closely resemble those of neighboring farms, feeding what they produce into bulk collection points (grain elevators, for example) and from those into processing, storage, and distribution networks that are optimized for certain measures of efficiency, notably the profitability of the companies involved.

But specialization of this sort does not lend itself well to high biological diversity, which, at the microbial level, is an important measure of soil health. This is less true of grazing operations, since pasture can be biologically very diverse, but even ranchers may suppress the growth of broadleaf plants to increase the proportion of grass in the mix, and, in any case, in temperate climates grazing may not be available all year around. For part of the year livestock may need to be fed hay, silage, grain, soy, etc., grown for this purpose or purchased from distribution chains similar to those described above.

So, even if we were already in possession of technology adequate to perform the detailed, plant-by-plant, leaf-by-leaf manipulations necessary to conduct diverse perennial polyculture at scale, we would still be faced with the challenge of pairing a production system with extended, overlapping harvests of perhaps dozens of different commodities from any given locale, with a processing and distribution system accustomed to large volumes of single commodities arriving more-or-less all at once.

This is both an informational challenge and also one of physical logistics.

The informational component begins with estimations of how much of what is likely to available from whom, how soon, with what expectation of quality. This can be combined over multiple farms to aggregate marketable quantities of each crop. By the time harvest arrives how much of each farmer's production should be combined with that from which other farms to go where should already be worked out.

The physical component is an exercise in materials handling, beginning with harvesting particular plants or plant parts from the midst of dense, mixed stands, without damage either to the crop being harvested or to what is left behind after harvest. The next step will likely be some form of containerization, so each farm's output can be identified, even after being combined with that from other farms for transport.

The informational component isn't so different from other applications of information technology, and, given the ability to provide reliable harvest predictions, developing that will be relatively straightforward. Most of what has yet to be worked out relates to the physical, materials handling component, the details of how to perform delicate harvesting operations and how to handle the crops once they're harvested.

Since efforts to develop such capabilities can't be expected to turn a profit next year, perhaps not even within the next ten years, this is an appropriate focus for government-funded research, and we should all be insisting that the relevant government agencies make this type of research a priority, while some soil remains to be saved and regenerated.

Friday, February 23, 2018

The point isn't to replace people; it's to go beyond what people can do at scale.

Imagine a horticulturist at work, repotting plants, snipping off withered leaves, moving plants farther apart as they grow, setting them out in gardens and landscapes, diagnosing nutrient deficiencies, diseases, infestations, and so forth, maybe even engaging in a little selective cross-pollination to produce new varieties, as time allows. A horticulturist needs to be a soil scientist, a botanist, a plant pathologist, a designer, and, in many cases, an entrepreneur.

Now imagine automating all of the most tedious physical tasks.

Even that is no small feat; skilled horticulturists probably have among the most secure jobs in existence. Nevertheless, it can be done and will almost certainly eventually be done.

So now imagine that you're running a company that is closing in on that goal and looking forward to what to take on next. Do you attempt to automate those aspects of the work requiring design and/or business sense, to entirely replace horticulturists with robots? Or do you press on with tackling more delicate tasks, faster, based on increasingly comprehensive sensory data backed by increasingly sophisticated processing and accumulated knowledge, with the intention of gradually moving this approach to plant cultivation out onto broad acres as the cost of doing so falls far enough to be competitive with other methods?

I suppose the answer depends on who you perceive the customer to be and what you perceive to be marketable, and whether you're only interested in business-to-business relationships or also want to serve the end consumer. (Note: people may quite willingly buy their food from a vending machine, but there's likely to be much more resistance to the idea of buying potted plants or flower arrangements from one. The interaction with another human being is integral to such transactions, and for good reason.)

I see the role of robots as coming in stages, beginning with augmenting what skilled humans can do by relieving them of the most time-consuming, tedious tasks, then taking the performance of those tasks deeper, by improving precision, data collection, and the conversion of that data into knowledge, then moving the capabilities developed into a broader context, such as by applying horticultural methods at the scale of agriculture.

The goal shouldn't be to replace humans, particularly not skilled, knowledgable humans. It should be to displace crude methods by bringing skill and knowledge to bear at scale.

Friday, December 29, 2017

Terminology: Replacing 'mechanical' with 'physical'

I have repeatedly referred to "mechanical methods" (of weed control, for example), both here and elsewhere. However, it has occurred to me that this phrase is far too constrained, failing to invite the broad-ranging imagination that will be needed, going forward. My sincere apologies.

'Mechanical' implies rigid structural components in more-or-less direct contact with whatever is being manipulated. Even solid components with a high degree of elasticity, such as are frequently used in soft robotics, aren't clearly included, much less devices with no moving parts, like phased-array radars, that project energy in a controlled manner.

"Physical methods" is a better, more inclusive term. The point is to differentiate methods involving pressure, momentum, impact, heat, cold, acoustic disruption, and so forth from methods involving the application of toxins.

"Physical methods other than tillage" is even better, since it helps avoid confusion with conventional physical methods. We need physical methods compatible with perennial polycultures which can be highly automated so they can be applied at scale. To this end, terminology which blinders the imagination is not helpful.

Friday, April 07, 2017

Crash Course Computer Science video series

Crash Course is a video channel you can access through YouTube.

Crash Course Computer Science is an ongoing playlist of fast-moving videos that begin with the most basic underpinnings of digital computing and progress from there.

If you are a complete novice, you'll probably want to take them one at a time, watch each several times, and perhaps also supplement them with other sources, to make sure you understand each new concept.

If you're already somewhat familiar with the subject, this is a great review!

Monday, March 13, 2017

What's wrong with ‘best practices’?

What's meant by ‘best practices’ and what could possibly be wrong with it? (Yes, I've used that phrase frequently myself, but I've been becoming increasingly uncomfortable with it.)

So I take ‘best practices’ to mean, of the practices now in use, those which come closest to achieving commonly agreed goals. This begs the question ‘commonly agreed among whom?’

In the context of agriculture, depending who you ask, goals that are considered to be ‘commonly agreed’ might or might not include such fundamentals as soil conservation, crop heterozygosity (in-species genetic diversity), and eliminating the use of broad-spectrum toxins, and might also include maximizing the growth and profit of certain corporations, as well as maximizing agriculture's contribution to alleviating the imbalance in imports versus exports.

Without going into detail, what constitutes ‘best practices’ can look very different depending on the goals to be served, but even leaving that aside there's still the issue of ‘best practices’ implying that the matter is already decided.

There's also a problem with ‘now in use’ as used above. As our tools and understanding evolve, what is now in use is almost certain to be replaced with something better, by some measure, eventually if not sooner. By focusing on even the best of what is now in use, we may miss the opportunity to make further improvements sooner rather than later, and, in the context of eroding soils and evaporating genetic diversity, in time to prevent further damage.

The focus of this blog is on the application of robotics to (addressing the many problems with) agriculture. I understand this in not obvious to many, both roboticists and high-concept gardeners and farmers (those engaging in organic / biological / biodynamic / regenerative / natural systems ... practices).

I don't see robotics as a magic bullet. It could all too easily simply accelerate the damage, in fact that's probably the default, in the absence of advocacy for making the technology serve higher aspirations, and without funding for achieving them.

What I do see is a deep well of potential for improvement, in tools, in the practices those tools enable, in our understanding of how and why to wield them, and, most importantly, in the scale at which those new tools, practices, and understanding can be applied.

Robots are machines that make decisions, sometimes very simple decisions but decisions nonetheless. They make those decisions on the basis of information gathered from their environments, and vary their behavior accordingly.

They are very good at tracking a multiplicity of details, and can learn from experience and carry out experiments. And, while they were until recently clumsy and slow, they are now gaining delicate touch, dexterity, and speed.

The potential I see is for replacing broad-acre monoculture with something better, much better, on the scale of millions of acres, but for that to happen in a reasonable time-frame developing the technology must become a higher priority than protecting currently profitable operations and arrangements, and so far the demand for this is both unfocused and lackluster. I'm doing what little I can to change that.

Check out The Permaculture Podcast, #1709: Regenerative Agriculture with Ethan Roland!

Sunday, March 12, 2017

Collaborating machines and avoiding soil compression

Soil compression can be a serous problem, but it isn't always and in all ways a bad thing. For example, the impressions made by hoofed animals, so long as they only cover a minor fraction of the soil surface, create spaces in which water can accumulate and help it percolate into the soil more effectively, avoiding erosive runoff.

The linear depressions made by wheels rolling across the surface are more problematic, because they create channels that can accelerate the concentration of what would otherwise be evenly distributed rainfall, turning it into a destructive force. This is far less serious when those wheels follow the contour of the land rather than running up and down slopes.

Taking this one step further, if it is possible for wheeled machines to always follow the same tracks, the compression is localized and the majority of the land area remains unaffected. If those tracks are filled with some material though which water can percolate but which impedes the accumulation of energy in downhill flows, the damage is limited to the sacrifice of the portion of the overall land area dedicated to those tracks and the creation of compression zones beneath them, which may result in boggy conditions on the uphill sides of the tracks, which may or may not be a bad thing, depending on what one is trying to grow there.

(I should note at this point that such tracks, when they run on the contour, are reminiscent of the ‘swales’ used in permaculture and regenerative agriculture.)

Tractors with GPS guidance are capable of running their wheels over the same tracks with each pass, but the need for traction, so they can apply towing force to implements running through the soil, means that those tracks will constitute a significant percentage of the overall area. Machines, such as dedicated sprayers, with narrower wheels that can be spread more widely apart, create tracks which occupy far less of the total land area, but they are not built for traction, and using them in place of tractors for all field operations would require a very different approach to farming.

It is possible to get away from machine-caused soil compression altogether, using either aerial machines (drones) or machines which are supported by or suspended from fixed structures, like posts or rails.

Small drones are much like hummingbirds in that they create very little disturbance, but they are also limited in the types of operations they can perform by their inability to carry much weight or exert significant force. They're fine for pollination but you wouldn't be able to use them to uproot weeds with tenacious roots or to harvest watermelons or pumpkins.

On the other hand, fixed structures and the machines that are supported by or suspended from them have a significant up-front cost. In the case of equipment suspended from beams or gantries spanning between rails and supported from wheeled trucks which are themselves supported by rails, there is a tradeoff between the spacing of the rails and the strength/stiffness required in the gantry. Center-pivot arrangements also have such a tradeoff, but they use a central pivot in place of one rail (or wheel track), and it's common for them to have several points of support spaced along the beam, requiring several concentric rails or wheel tracks.

Strictly speaking, there's no particular advantage in having rail-based systems follow the contour of the land, since they leave no tracks at all. Center-pivot systems using wheels that run directly on the soil rather than rail are best used on nearly flat ground, since their round tracks necessarily run downhill over part of their circumference. In any rail-based system, the ‘rail’ might be part of the mobile unit rather than part of the fixed infrastructure, drawing support from posts spaced closely enough that there were always at least two beneath it, however this would preclude using trough-shaped rails to deliver water for irrigation.

Since the time of expensive machines is precious, it's best to avoid burdening them with operations that can be handled by small, inexpensive drones, and the ideal arrangement is probably a combination of small drones, a smaller number of larger drones with some carrying capacity, light on-ground devices that put little pressure on the soil, and more substantial machines supported or suspended from fixed infrastructure, whether rail, center-pivot, or something else. Livestock (chickens, for example), outfitted with light wearable devices, might also be part of the mix. In addition to gathering data, those devices might be used to direct their attention and to defend them from predators.

The small drones, being more numerous, will be the best source of raw data, which can be used to optimize the operation of the larger drones, on-ground devices, and the machines mounted on fixed infrastructure, although too much centralized control would not be efficient. Each device should be capable of continuing to do useful work even when it loses network connection, and peer-to-peer connections will be more appropriate than running everything through a central hub in some circumstances.

This is essentially a problem in complex swarm engineering, complex because of the variety of devices involved. Solving it in a way that creates a multi-device platform capable of following rules, carrying out plans, and recognizing anomalous conditions is the all-important first step in enabling the kind of robotics that can then go on to enable scalable regenerative practices in farming (and land management in general).