a gardening robot

I like it. I, personally, hate gardening and would love to have a robot plant flowers for me.

I use two 12V SLAs in serial for 24V. This is run through a switching regulator to convert it to 12V. Using deep discharge batteries I have a 4-5 hour run-time for my electronics. The motors run on a separate battery.

I was originally going to use AA NiMHs, but I had major problems and quickly gave up on that. However, I'm using a fairly high- current set of electronics (a Pentium-M Mini-ITX board).

-- D. Jay Newman

What OS and programming language are you planning to use?

John

The OS is simple: Linux! I have Fedora Core 4 (test 3) installed on the machine already. Gentoo failed to boot for some reason.

As for the programming language - I'll be starting off with PHP+cron, but converting that to C as I become more confidant that the programs are correct. I've been thinking about this project for a very long time, and have quite a few ideas on how to write the code (I'm a programmer by career). The hardest part, in my opinion, is actually building the robot, as I have very little experience with electronics.

Kae

My main reason for building the robot is to eventually have an automatic vegetable garden. I think aesthetic purposes may be a bit far down the road for the robot yet! The closest I think it would come to ornamental gardening is in keeping the garden clean, trimming the grass, seeding patchy grass areas, and trimming hedges (if I make telescopic arms). I'm mainly interested in having my bot keep patches of vegetables weed free, and eventually, hopefully plant and harvest the vegetables as well.

Perfect. Thanks a lot. I'll check up prices and go work on that knowledge.

Kae

I'd have thought that the hardest part was reliably spotting which plants were actually weeds !

Possible Algorithms ?

Dave

Posted Via Nuthinbutnews.Com Premium Usenet Newsgroup Services

---------------------------------------------------------- ** SPEED ** RETENTION ** COMPLETION ** ANONYMITY **

----------------------------------------------------------

lol! yes, but that algorithm might also zap my 2-year-old kid.

I'm working on a connectionist network for reliably recognising various different plants. The bot will only attack what it is very sure is a weed. I'll make the code freely available once I'm certain it's working the way I want it to.

a quick example of connectionist networks:

there's a style of gardening that ignores what some would consider weeds, just keeping them short right near the actual crops will do the same thing as removing them. to get the plants started the weeds are kept back, other than that things are left natural, and apparently the crop quality increases. some others are more extreme, the wild grasses are allowed to grow regardless, you walk into the fields and accidentally step on a watermelon, you get the idea what's different about that type of gardening.

true. In fact, it would easier to get a bot to just trim a weed anyway, as digging up roots could get messy. Of course, it means the action would need to be repeated much more often, but then, it's not like the robot will get bored of doing it, right?

to do with allowing the garden to go as it wants - there is an extreme form of gardening I was interested in for a while, where you get a load of different seeds of things you want to grow, and you simply mix them all up together with some wet clay to form small "seed balls", then simply throw them wherever you want in the garden. The appropriate plants for the micro-climate in that area of your garden will grow.

but - that's off topic ;)

Back to the bot - I'm currently playing with an old printer cable. I've stripped the shielding off the printer end, and am having fun trying to figure out how to send a voltage down the wires. Here's a page I'm finding useful for that:

One thing I never figured out is how the standard neural net recognizes a feature no matter where it is in the image and also output that location?

In the example site you gave they mention recognizing the wearing of sunglasses and how the biggest change in the connections is around where the eyes are most often located versus other parts of the image. So how do the connections with the biggest change find the eye locations in the first place?

One example of a nets limitations was the military training a net with a series of photos some with tanks in them and some without. What they didn't realize was the images with the tanks were taken on an overcast day and that is what the net learnt to recognize!

The easiest way to kill a weed would be a squirt of weedicide, hot steam or with high voltage electricity. In broad acre farming there was a system that used the green color to just squirt the weeds not the bare ground. In another system the problem was very high weeds. An electric wire at the right height zapped the weeds leaving the smaller crop below untouched.

Disturbing the soil by tilling can actually give dormant weed seeds a trigger to grow.

John

well, my own net won't be saying where in the image the weed is - just whether or not the weed appears in the image. That will reduce the complexity of the net. Then, by testing different parts of the same image, we can get the location of the weed.

that's a difficult question. the point of neural nets is that you don't tell them how to solve the problem - you don't say "check around the eyes", etc. Instead, you provide a load of test data with their solutions, then you test the net against the data, "correcting" the net when it gets it wrong, and "rewarding". I think of it as "carrot and stick training".

lol! yes. I'd heard of that one. Well, that's an example of the trainers making a stupid mistake.

true - like stirring a puddle can raise some silt.

The method I'll be using is to snip the weed just above the ground, and compost the waste. The root will eventually give up after continual abuse in this way.

/me goes back to trying to get the wireless network card working in his bot Kae

I wouldn't count on that; it depends on the plant. One of my neighbours was complaining about this kind of weed that keeps coming up no matter how often he trims it, mows it, etc. etc. He learned from an expert that this particular plant propagates underground, and its roots can go as deep as seven feet! He lives beside woods where this weed flourishes. There's no way to dig it up, and no way to starve it. I've tried the same thing with dandilions and haven't been able to make it work. They're more persistent than I am. (Maybe a robot will be, too.)

But you're probably going to be okay for 99.99% of what you want. I'm really hoping your project will go well.

- Owen -

Kae Verens wrote in news:g53Ce.2556$ snipped-for-privacy@news.indigo.ie:

To spray them buggers with herbicide seems to be the easiest solution. But, if you don't like the idea of flinging poison around, then you need something else.

How about:

Attach a "sampling lab", similar to what some of the Mars missions have/ had. The 'bot snips a leaf, puts it in a digester , analyzes it and takes appropriate action based on the chemical makeup of the specimen. This way, you could cultivate and fertilize simultaneously.

I'd recommend some sort of flame thrower to kill the weeds. Great nightime entertainment!

Joe

woo! too hi-tech for what I was thinking!

lol! I had a thought once of a laser-cannon that I wanted to build, which would zap flies, filming each execution, then on weekends, I could play the videos while having a nice cool drink.

Kae

Thank you. The more I think of it, the simpler the task seems, and the more puzzled I am that no-one else has done this - after all, if it really works, then I will essentially get free food for the price of some electricity and minor maintenance; and electricty can be provided free as well, when I have the time to build up some wind/solar energy collecters.

Kae

a lot of weeds can be effectively killed by hosing them down with hot water(it must becontained or the heat will bleed off too fast), melts the waxes on the surface and leaves it unprotected. for continuous weed killing you can run a bot down a row towing a small sheet of heavy rubberised cloth, spraying hot water underneath, then the problem becomes storing enough water and keeping it hot.

Actually, I think there is someone in Alberta doing research for agriculture. They're using (or plan to use) differential GPS to know the position of every single seed they plant in an entire field, and they want to have robots remove every plant that isn't at one of the expected positions. This info is second hand, told to me by a friend, so take it with a grain of salt (or wheat). But that's the basic idea and it seems like a good one. What do the robots care if it's a tedious lot of work, eh?

- Owen -