Open source project proposal - call for contributers


#1

Dear openAG community,

I’m looking into ways to contribute from my knowledge to the domain of Precision Agriculture and came up with an idea (though I’m probably far from being the first) for holistic project that I believe can make a great impact.

After speaking with many professionals in related domains, and reading many (many) papers on the subject,
I’ve put down the idea in broad terms into a presentation page:

https://hemulin.github.io/auri/

Would love to hear your thoughts on that matter.

How would you recommend me to move forward?

Any feedback would be highly appreciated.

Best,
Yotam


#2

Hi there,

Two Quick question with respect to that project :point_up_2::

As I understand it, the food computer is a fully controlled climate environment that can provide “noise-less” or least reduce noise of external factors while growing plant.

  1. How does the plant phenotype is being measured? Are the recipes based on a “feeling” (e.g - I’ve grown extra sweet tomatoes using that recipe) or is there a standard way by which nerdfarmers are expected to measure their plant expression? If so, what is that measurements standard

  2. I thought of a missing link and I’m not sure if I’m right -
    I think on setting up a food computer (which is clearly a very small scale production), placing, within the growing chamber, an off-the-shelf multi-sensor device (perhaps the Libelium Waspmote), try out various recipes and then binding between their success (outcome expression) to the set of measured attributes by the multi-sensor.
    Doing so will enable taking that multi-sensor to a wider scale production, having the same set of readings that was mapped in the small scale, and extrapolate it to a large scale production. Where the goal would be to replicate the food-computer yield to the large field yield by minimizing the gap between the food-computer readings and the field readings of the same attributes.

Does it sound logic? Am I missing something?
(@webbhm, you seem to be a good address for this question so I’m tagging you here)

Cheers.


#3

The MVP has some environmental control, but it is not total control - there is no cooling or control of gases (ie CO2, humidity). We are primarily focusing on light and temperature. The MIT Food Computer is going for a higher level of control, but at a significantly higher cost.
Phenotype is an unending subject. I retired from Monsanto, where it was an issue just to define and track the names of the phenotype attributes that were measured. This is probably our weakest area at the moment, with the recording being manual entry to spreadsheets. We are trying to tag along with Fairchild Gardens and the data they collect, which is mostly plant dimensions and harvest weight. Just getting that is an effort at this time. We are going slow here, because most people seem to be interested in just growing plants, and not into hard-core research. The main focus has been on improving plant growth (time and volume), though there are a lot of other areas to explore (light, CO2, nutrients, flavor).
Yes, something like the Waspmote is great for all it collects and would be ideal; but to keep within our price range we have limited the sensors. I don’t know about the Waspmote, but we have found that a lot of the ‘packaged’ sensors are difficult to integrate into a larger system as they don’t have open standards and are usually a closed, proprietary software. However, you larger concept of running recipes and comparing results is what we want to do. To be honest, sensor data is the least interesting and useful data. If you correctly follow a recipe, the environmental data only tells you that you followed the recipe - it is the phenotype data mapped to the recipe that is the real value.
You are right on target that the long term goal is to do research on a small scale, then see if the results can be applied to larg scale commercial systems.
I don’t think you are missing anything, you understand our goals. Our real challenge is with the economics and getting the discipline of consistently collecting phenotype data.
At the moment I see us as having two primary goals: 1) Developing a cloud UI where people can share data and input phenotype observations, and 2) defining a data standard for collecting sensor and phenotype data that will be useful for operational, administrative and analytic needs (we have a model, the challenge is getting community buy-in).
Thanks for your interest and reaching out to us.


#4

Thanks for the quick reply.

We are primarily focusing on light and temperature

My small research suggests that there are many (many) attributes that effects the plant phenotype, some to a small degree and some have a very big impact. Why not control/measure those as well (NPK values for a start)?

To be honest, sensor data is the least interesting and useful data

Assuming hypothetically that there is cheap, open software multi sensor.
I have the MVP and I have second environment with control over irrigation and over nutrition.
Now I’ve finished an experiment with a recipe (which consisted of temp/light control) in the MVP, and during the experiment I’ve collected whatever data I could using that said sensor. wouldn’t I be able to replicate the MVP outcome in a different environment by mimicking and trying to minimize the “edit distance” of the same readings between the two environments?

Developing a cloud UI where people can share data and input phenotype observations

Does this task has clear specifications? Can I help in anyway in defining/building it?

we have a model, the challenge is getting community buy-in

You mean the community input of data into the model? If so, I guess it is something that this cloud UI can be a turning point for.
Is there in the software an activity that when an experiment has been concluded, automatically a form opens on the UI, asks the user, over a form, for a list of phenotype expressions and then submits it the cloud with that experiment recipe data?


#5
  1. Yes, there are many factors, but before we deal with the small degree ones, we want to get a handle on the big ones. As for NPK, if you know your fertilizer, you know what is going in and don’t need to constantly monitor the current nutrient state. This is also where some things are easier to measure manually (ie pH) than to try and automate it. We also see that some environmental measurements will need to be manually entered.

  2. Totally agree.However I would see both environments being measured for their fidelity to the recipe.

  3. The initial goals of the UI are data display (charting), data input and administration (validating users, letting teachers manage student access). After that would be recipe creation, modification, download and creating data summaries. I would envision interactions with the UI when:
    a) Someone registers a MVP with the system.
    b) Starting an experiment (download a recipe, register the start of the experiment)
    c) Logging manual data during the experiment and at harvest
    d) concluding the experiment (final data entry and summary creation)

fyi: @Webb.Peter is the main person behind the MVP, his idea and the main driver. I am more on the technical side, writing software and experimenting with sensors.


#6

before we deal with the small degree ones, we want to get a handle on the big ones

Fair enough. Can you name the first few top candidates on that list of “big ones”?

As for NPK, if you know your fertilizer, you know what is going in and don’t need to constantly monitor the current nutrient state

Agree, I’m facing a situation where my main target in the long run is to help Organic Agriculture, probably where the adjustments would consist of applying (for example) Compost contains high values of N over another Compost which has high value of K. In that situation, I can’t actually tell how much went in, or at least not in a very precise manner, that is why I would need those NPK measurements.

some environmental measurements will need to be manually entered

That is an interesting idea, I haven’t thought of “half-automated” control system. I see the merit of it (and some complications as well).
Can you name few of the measurements that are assumed to be manually entered (beside pH)?

MVP Cloud UI:
I get the gist of how you envision that application. Did anyone started working on it already? If yes, I’d love to join in. If not, I’d be happy to setup some prototype to iterate on.

Thanks for tagging @Webb.Peter, I’ll drop him a note to see what his take on that cloud app.

In addition, I had a short correspondence with Rob Baynes from openAG and he said that they have a cloud solution and open phoneme database, but both are still being built and are not public yet. According to him they will be released to the community later this year.
Do you know anything about it? Structure/specifications/models/ETA for release?


#7

hemulin https://forum.openag.media.mit.edu/u/hemulin
June 18

before we deal with the small degree ones, we want to get a handle on the
big ones

Fair enough. Can you name the first few top candidates on that list of
“big ones”?

Light: photoperiod and spectrum
Nutrients
Temperature
For some good background, see the Cornell Lettuce Handbook
http://www.cornellcea.com/attachments/Cornell%20CEA%20Lettuce%20Handbook%20.pdf.
The basics are the same for all plants, though the details are species
specific. Vegitative growth is simple compared to dealing with flowering.
There is a lot of research on this with cannabis.

As for NPK, if you know your fertilizer, you know what is going in and
don’t need to constantly monitor the current nutrient state

Agree, I’m facing a situation where my main target in the long run is to
help Organic Agriculture, probably where the adjustments would consist of
applying (for example) Compost contains high values of N over another
Compost which has high value of K. In that situation, I can’t actually tell
how much went in, or at least not in a very precise manner, that is why I
would need those NPK measurements…a

With a lot of organics, the issue is you don’t know the chemistry of the
fertilizer You may want to consider a garden soil chemistry kit to check
the compost. I know of some commercial soil testing companies that will do
it for about $15/sample, but a simple kit from a garden center will get you
in the ball park to start.

some environmental measurements will need to be manually entered

That is an interesting idea, I haven’t thought of “half-automated” control
system. I see the merit of it (and some complications as well).
Can you name few of the measurements that are assumed to be manually
entered (beside pH)?

Most of the chemistry (NKP) would be manual at this time. The issue is
cost, it is cheaper to do a lot of this manually than the cost of the
sensors (and issues with calibration). All most all of the phenotypic data
is manual, though we are looking at trying to get plant size from images
(see the Danforth Center’s PlantCV
https://github.com/danforthcenter/plantcv for some good work).

MVP Cloud UI:

I get the gist of how you envision that application. Did anyone started
working on it already? If yes, I’d love to join in. If not, I’d be happy to
setup some prototype to iterate on.

Thanks for tagging @Webb.Peter
https://forum.openag.media.mit.edu/u/webb.peter, I’ll drop him a note
to see what his take on that cloud app.

I would suggest you contact Peter, and see if he will get you into the
development forum where you can contact the developers. Right now we are
exploring different libraries and packages (Polymer, ThingBoard, Blynk, …)

In addition, I had a short correspondence with Rob Baynes from openAG and

he said that they have a cloud solution and open phoneme database, but both
are still being built and are not public yet. According to him they will be
released to the community later this year.
Do you know anything about it? Structure/specifications/models/ETA for
release?

I have talked with Rob at various times in the past, but find that he is so
overwhelmed with tasks that he hasn’t had much time to share or collaborate
on the work.


#8

Hi, sorry for the late reply.

Thank you dearly for that post. It greatly helps in pointing me in to the right directions.

The cornell guide and the phenotype image processing library are very helpful.

I think that at this point, and with your latest addition to the relevant-links-collection I have much of the information I need to just get down to business and start building stuff.

The way I see it, I’d just setup growing environment (small area outdoor, small area indoor - perhaps greenhouse simulator, and PFC) and start running as many experiments as I can with control groups.
Collecting on the way every metric I can find (now image processing is also in the pipeline), and map readings to phenotype outcome.

Once the Open Phenome DB is out, I’ll add my data to it and see if there are any clear correlations between my findings and others.

Are there any types of crops you would recommend to start with? Perhaps something which presents a large degree of phenotypic plasticity as well as popularity among other PFC owners so I can find other result to compare against.

p.s
I couldn’t find out how to send PM to users in this forum, I sent Peter a message on FB but perhaps he missed it.


#9

To ping someone directly you do a @hemulin, That alerts them directly through the forum.

Yes, in theory you set up as many experiments as you can and collect every metric. In practice it is better to start small and get everything running smoothly. I would suggest identifying several metrics you are interested in, and focus on them to start. Most people are starting with leafy greens (lettuce) and getting the basic metrics of size and harvest weight. When you get into flowering and fruiting, things can get complex fast. Find what interests you and follow it, though most of the research focuses around crops that have an economic interest.


#10

Much appreciated.

Will post important updates once I have them.

Thanks for your help and your great work :slight_smile: