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).

Monday, January 23, 2017

New blog title: Regenerative AgRobotics

The title of this blog until today, Cultibotics, has always been kind of awkward, both invoking unfortunate associations (cults) and failing to communicate its purpose without explanation.

I hope the new title, Regenerative AgRobotics, will serve better.

For compatibility with existing links, the URL remains https://cultibotics.blogspot.com. (I will endeavor to remember to use the secure https version henceforth.)

Wednesday, January 18, 2017

How Regenerative Agriculture and Robotics can benefit each other

I've come around to the view that the best and most inclusive term for high-concept farming which is both sustainably productive and ecologically responsible is Regenerative Agriculture. It implies all that is meant by permaculture, agroecology, carbon farming, and organic farming, but goes beyond these to focus on living matter in the soil, and in this is closely aligned with the term biodynamic. That said, I'm not prepared to argue the point; I only say this by way of explaining why I've chosen to use this term here.

These are perilous times, perhaps not to the extent portrayed by tabloid journalism, or suggested by the recent abundance of misinformation and misinformed opinion on social media, but perilous nonetheless, and there are plenty of squeaky wheels around all competing for grease. In this political-economic environment, making a case for adequate funding (whether public or private) has become increasingly difficult. The cost-benefit ratio or colloquially ‘bang for the buck’ is again the primary metric driving investment, with benefit usually measured in terms of potential profit where private investment is concerned.

The synergy of substantive collaboration (or ‘co-leverage’) can be all-important in determining which projects get funded. If you can show that advances in your field are synergistic with advances in another field, such that each furthers the goals of the other, and your proposal crosses those boundaries, you stand a better chance of attracting investment than if you go it alone.

I believe such a synergy exists, or could easily be made to exist, between robotics and regenerative agriculture.

Robotics needs a market into which it can sell a large volume of many kinds of machinery, and from which it can gain abundant experience, without the burden of liability presented, for example, by autonomous vehicles operating on streets and highways. Agriculture is good for this, in terms of scale, tolerance for beta-ware, and the opportunity for gaining experience with new approaches and new technologies, and operating within complex environments. Regenerative agriculture is particularly ideal, because the need is for small-scale, context-sensitive equipment, which, if it malfunctions, will only cause minimal damage, and the complexity of the environments involved is at least an order of magnitude greater than with conventional agriculture.

Regenerative agriculture needs a way of scaling up its best practices, so they can be deployed across millions of hectares, without resorting to compromises involving heavy equipment and soil compression, or herbicides, and without waiting for millions of people to decide they want to go back to the land and work it by hand. Because many of those best practices involve attention to detail, that scalability can only be achieved through robotics. On the other hand, given equipment that automates not only the work of regenerative agriculture but data collection, the stage would be set for pushing the state of the art, to evolve even better practices and to further develop crop genomes based on phenotype (how plants actually perform under diverse conditions).

Together, robotics and regenerative agriculture can do much to drive each other's development, and to improve humankind's prospects for the future. But, of course, the near-term bottom line is that, together, they can make a more compelling case for adequate funding.

Sunday, January 15, 2017

How technology can help keep farmers relevant

I've long believed that Augmented Reality (AR) and robotics are closely related. Both model their environments to some degree. Robotics uses that model to guide the behavior of a machine, whereas AR uses it to provide an enhanced sensory experience to a human.

The exact nature of that enhanced experience is bounded only by available sensory, computational, and display (audio, haptic, ...) hardware, and by how the data gathered can be usefully transformed into overlays that augment the natural perception of the human user. What is useful is a function of both the content of those overlays and the latency, how much lag time is introduced by the computations involved in generating the overlays. Faster computational hardware can produce more detailed overlays with the same latency or the same overlays with lower latency than slower hardware.

One important application for AR is making it safer and easier for a human to work in collaboration with robotic hardware. For example, a robot might provide the path it intends to follow and the 3D space through which it intends to pass, and that information might be converted in an AR display into highlighting of anything occupying that space. Or perhaps a machine wants to direct the attention of its human counterpart to some particular element of the environment, say one specific plant. That too could be highlighted in the display.

While these examples only scratch the surface of what is possible, they do serve to illustrate that the content of the AR overlays need not be generated entirely from data gathered by sensors attached to the display itself, but can be provided by other sources, including but not limited to other nearby devices. Those sources might include aerial or satellite imagery and information from databases. In the farming context, they might include 3D soil maps produced from core samples.

Examples of overlays that might be useful for a farmer include thermal imagery, current soil moisture content, soil surface porosity and water absorption capacity, exaggerated vertical relief and what to expect in the way of runoff and resulting erosion for various precipitation scenarios, highlighting all plants of a particular species, all plants exhibiting nutrient deficiencies or other trauma, highlighting bare soil (no mulch or plant cover), the presence, activity, and impact of various types of animals. This list could go on and on.

Machines may be better at doing particular manipulations of data, finding correlations, and even at answering well-specified questions, but they're not so good at asking meaningful questions, much less at thinking outside the box. For this reason, the combination of human and machine is more powerful than either alone.

It's still very early days in AR, and there's a great deal of room for improvement. One development that is likely to occur sooner rather than later is voice operation, enabling hands-free control of the AR experience, including which overlays are active and how they are combined. With voice control, a farmer should be able to walk through a field, say what he wants to see, and make modifications to the plan controlling the robotic machinery that actually operates the farm, or issue commands for execution by the first available machine. For most, this will be a more intimate and far richer connection to their land than what they currently experience.

Saturday, January 14, 2017

What's left for farmers to do when machines ‘do it all’?

Let's assume, for a moment, that the vision I've laid out in this blog is ridiculously successful, and, over the next few decades, robotic devices take over all aspects of tending land and crops and handling material inputs and produce, and do it using increasingly sustainable (regenerative) practices that begin the process of retaining and enhancing biological diversity and reviving overworked soils. What's left for farmers to do? Will there even be a need for humans on farms?

Well, in that improbable scenario, once the process has run its course and there's no room left for qualitative improvement, and the machines are self-repairing and the algorithms are self-optimizing for constantly changing conditions, perhaps, strictly speaking, the answer would be "no, not really, not a need in the sense of an indispensable component." A farm so equipped might run itself for months or years at a time without human intervention.

There are two problems with this, however. First, that level of automation is unlikely to happen in the lifetime of anyone reading this, and for the foreseeable future there will be need for human involvement in design and management decisions, making use of their own senses in combination with data to drive improvements, as well as filling in the gaps in the perceptual and physical capabilities of the machines.

The other objection has no expiration date. It's about whether a human presence brings anything valuable to land husbandry, and ultimately whether people are allowed to forget where their food comes from, such that, if all of the machines were to suddenly shut down or break, no one would have a clue as to how to proceed. There will always be a need for humans who understand why the machines make the choices they do, and that understanding will always be best achieved when augmented by being physically present, walking fields, examining details with a magnifying glass, experiencing the sounds and smells directly, and lending a hand with some of the work, even when it could be done faster and more precisely by one of the machines, just to have the experience of participating in the process.

Without such people who remain deeply connected to the land and our use of it to act as interpreters, those who live in urban areas will have only feeds from the machines themselves and will probably lose interest, and perhaps also lose the will to maintain support for environmentally responsible policies in the face of periodic calls for austerity and ever-increasing efficiency. Actually, we're faced with that challenge right now, so maybe there are already too few people on the land, or those who are engaged in farming don't have the time for learning about all of the ‘extraneous’ factors not obviously relevant to their low-margin agribusiness operations, or for acting as interpreters for others.

Sunday, January 01, 2017

The coming age of AI: What is NOT inevitable

In a Backchannel post Steven Levy excerpts “the seven stages of robot replacement” from Kevin Kelly's book The Inevitable: Understanding the 12 technological forces that will shape our future, and briefly lays out the main thesis of the book, which is that AI-driven transformation is coming, whether we're ready for it or not. (Caveat, I have not read the book itself and cannot speak to what all it may or may not include.)

Artificial Intelligence is in the process of arriving, a process which began a few years ago and which is likely to take another few years or decades to play out, but that it will play out is inevitable, and that this will mean a degree of social and economic disruption is equally inevitable. What remains in question is how that arrival will effect everything else, which will be in no small part a function of the forces and specific decisions shaping the translation between machine perception and behavior.

If that translation is optimized for maximum profit for those funding the development, production, and marketing of these technologies, will it reflect concerns which don't rate as priorities for those who are their direct customers, those who buy and operate the machines, presumably hoping to optimize the profitability of their own businesses? These customers must operate in their respective markets, dependent on the demands of their own customers and the cooperation of other suppliers, each driven by their own interests. If their margins are thin, they may have no patience for machine behavior driven by what, for them, are secondary or tertiary concerns, which, in the context of agriculture, could very easily include soil health and erosion control, if catering to those concerns makes the difference between profit and loss in the current year.

AI and robotics could enable the scalability of ecologically sound practices in agriculture, but will they? There's more at play here than the inevitable march of technology. While the agricultural context is really only an example, it is one where concerns other than short-term profitability are all too likely to fail to make the cut as priorities. This would be less worrisome if those other concerns didn't also represent our best hope for halting the loss of biodiversity and soil fertility, for stemming the accumulation of greenhouse gases in the atmosphere, and for improving human nutrition.

AI is inevitable, but the most desirable benefits from it are not, and it will take more than market forces to ensure they don't remain forever beyond reach.