How to Detect Plants - Automatic Garden Weeder

I recently found "Tertill", a garden weeding Roomba-like robot. It uses the theory that weeds are short and plants are tall. When it finds a weed, it uses its mini weed-eater to chop it down. It uses "technology similar to your smartphone's buttons" to turn it away when it bumps into a large plant, and to turn on the mini weed-eater when it finds a weed.

I want to make my own version using an Arduino.

My Question- How to sense the plants. --- Big plants should be easy. (It needs to steer away from them.) Long-lever limit switches should work; I think Tertill uses capacitive sensors. Can a library like CapSense or ADCTouch sense plants?

The little weeds will be a lot harder. How would you sense when it drives over a weed? I'm hoping to do it with something cheaper than a Pixy camera.

I have a fair bit of coding experience, and would call myself an advanced Arduino programmer.

Thanks in advance!

Have you tried an IR setup? Perhaps there is enough reflectivity difference between dirt and plants to work with. What are you going to do if it gets loose in the lawn??

A sensitive IR sensor might detect that plants are a bit cooler than the soil via transpiration in the leaves…

never tried it.

Allan

Green plants reflect near-IR strongly. A true IR photograph will show all trees and plants as bright compared to soil, people or the sky.

Far-IR (black body or thermal radiation) is much less distinct - the plants will be very close to the same temperature as the soil.

Since most cheap IR sensors are near-IR, it's easy to buy a few different configurations and test them.

I'd build the thing with a physical bumper so it just can't drive over the top of taller plants. Then put the weedeater inside that perimeter and run it all the time.

Surely green plants are green because they aborb the blue and all the longer wavelength light - red and near IR?

Not my expertise.

But commercial glasshouse growers of lettuces (friends of mine) do use an optical thermometer to see how well their plants are transpiring.

Allan

Nope, no near-IR absorbance.

I suggest trying two-channel detection, using both a green and an IR LED in photodiode mode to detect light reflected from green plants (LEDs are very wavelength-sensitive in their response).

Cameras are out of the question with AVR-based Arduinos.

|500x320

Keep in mind that Kickstarter projects are entirely theoretical until delivery, and very often even after delivery. I strongly doubt the "Tertill" will live up to its hype.

tinman13kup:
What are you going to do if it gets loose in the lawn??

You need to put a short fence around the garden.

I’m thinking of trying IR. Are we looking at a IR led and receiver, a Sharp distance sensor, or something else?

Why bother? If you basically have a tiny lawn mower which stays away from larger plants then anything that gets under the blade should be bare ground or a smaller plant (or maybe a potato bug). So, why not just place the "weed whacker" underneath (inside the Perimeter with sensors which prevents running over big plants) and leave it on all the time?

Weed detection has been .... and probably still is... a challenging thing to do. Do properly that is. It probably requires camera and image processing technology .... as part of the system. Like.... identifying shape, colour, and things like that.

I'm also interested in reproducing one of these. Did you make any progress?

Swiftnesses:
Did you make any progress?

No, it’s one of those arduino projects that goes on the bottom of a to-do list; line number 306! :stuck_out_tongue: :stuck_out_tongue:

It is extremely difficult to "sense plants". I think that apart from having some fun with IR detection (that nevertheless shall be combined with UV detection) getting to a usable robot based on this concept is going to take long time. Range of the sensors and the multiple combinations of plants that can be found on site are major obstacles.

However, a robot to collect information about a certain vegetation plot is easier to build, in my opinion. It is on my list, too. In my case, the robot shall collect info, transmit or store it and then humans can review that info.

A sort of mobile sensing device, equipped by RGB sensor, IR sensor, humidity, temperature and light sensors.

Myself I work now on a Robo-cat. I will transfer part of his components and code will to that Robot. I am not going to design the Robot to do something on-site or to act on its own in some way. That requires far more sensing capacity, artificial intelligence and complex actuators than I have.

If you advance in your project, I will be happy to learn from your experience, if you will be so kind to share it here.

falexandru: It is extremely difficult to "sense plants".

Totally agree. A very interesting topic and field here. If they can get robots to be 'humans'.... or see and think and reason like humans....... then lots of progress could be made in this area. But then again, lots of unprogress would be made as well ..... due to the danger of having bots that are like humans (or are humans).....especially since some humans are a danger already.

Decades ago huge money were spent on so-called "artificial intelligence" - which proved at that time just a way to throw money away.

Today, there is a new "spring" of the artificial intelligence.

Perhaps the trigger is the advance in quantum computers, I do not know.

With regard to what a robot can do on plants, that is indeed an interesting topic. I am now experimenting with an RGB sensor. That can detect how "green" is a leaf, if it is in its range. That is the first problem: how to get the leaf in the range of the sensor in a reproductible way. One way is to go collect hunders of reading and then average them.

It is an RGB sensor a "remote" sensor? I really do not know. For me, if it can detect reliable up to 2 inch (5 cm) that is ok. "Remote" enough. That has to be proved - first on my desk and then in the wild.

Temperature reading is another tricky fact. Having a reliable reference is a must. Collecting and averaging is a must.

Humidity - that depends on temperature and some sensors do correct the reading accordingly. How accurate? I do not know.

Light - still no idea.

UV- stil no idea.

being one to toss in a nickel in when no one asked for my 2 cents, and comment on things of which I know nothing about.....

If this was presented to me, my first thought would be to use a much more powerful processor and see if one could do a video comparison. The camera / video comparison would seem like a good choice.

a low cost alternative is to have a guide and just weed-wack everything in the walkway. this does not address the space between plans. for that there is a ground cover or garden cloth.

or, move indoors, do hydroponics and enjoy the fish and salad every now and again.

dave-in-nj: being one to toss in a nickel in when no one asked for my 2 cents, and comment on things of which I know nothing about.....

If this was presented to me, my first thought would be to use a much more powerful processor and see if one could do a video comparison. The camera / video comparison would seem like a good choice.

a low cost alternative is to have a guide and just weed-wack everything in the walkway. this does not address the space between plans. for that there is a ground cover or garden cloth.

or, move indoors, do hydroponics and enjoy the fish and salad every now and again.

May I also add to this discussion. If the OP and others had spent much time raising a garden, they would know that weeds are not removed by cutting them off. that only encourages them! You actually have to pull the top and the roots from the ground to kill the things.

Paul

Paul_KD7HB: May I also add to this discussion. If the OP and others had spent much time raising a garden, they would know that weeds are not removed by cutting them off. that only encourages them! You actually have to pull the top and the roots from the ground to kill the things.

Paul

Around my area, soil compaction is the problem, and the garden weasel takes care of that along with any weeds in the way.

Perhaps using a similar setup to drive the bot could eliminate the need for the trimmer and solve compaction issues as well. Then it wouldn't need to sense any weeds, only taller plants to NOT run over.

Paul_KD7HB: May I also add to this discussion. If the OP and others had spent much time raising a garden, they would know that weeds are not removed by cutting them off. that only encourages them! You actually have to pull the top and the roots from the ground to kill the things.

Paul

Exactly. And some weeds have nodules in the ground, where those nodules just stay there when the plant gets pulled out. So, for some weeds, the human-like artificial intelligence machine should recognise the plant, and then somehow use a pesticide to touch the plant in various spots, without dripping onto other plants.

This of course is written out of respect to plants....as it's not their fault that they're living. Just trying to survive like everything else on the planet.

Baby steps. :-).

If the bot could get spectral data (say RGB spectrum, perhaps some IR and UV) about plants in its close proximity (sort of speaking) then we have a sound input for any further action one may choose to perform.

I attempted to use regular RGB to sense colors. In theory, looks fine. However, the R and G and B values are so close that is really difficult to interpret them.

So, how to get reliable and useful data about plants, using the regular Arduino-compatible components?

And I am speaking from the perspective of the one who just wired up the RGB sensor on both UNO and NANO three time by now, using 2 codes. And still no success.