UPDATE: GAGA team hunts down grass-smoking ROBOT Poor weather has played havoc with our blade-testing, so the Genuinely Autonomous Garden Assistant (GAGA) team has been focusing on location-tracking technologies, with near-uniformly disappointing results. Who would have believed that in the first years of the 21st century, a robot going about its business could not be …
Try using two IR LED's on your stick (top and bottom), then put an IR Trans filter on the camera Job done. You now can only see the LEDS not background and you can measure the height (Distance between).
You could use a third, Mid-height, LED to code which stick it is looking at. (ratio of gap between mid and top:bottom LEDS.)
Not sonar - triangulation based on ultrasound. Right frequency, wrong choice of transmitter/receiver.
Just put a source at each corner of the lawn, modulate with an ID and timing (you can wire them to a central source) and read the sources at 3 mikes each a foot or so from the center of the lawnmower.
The rest is a trivial computation problem - you should have no problem nailing the coordinates and orientation of your lawnmower down to under 1m. Add some roomba style bumpers to prevent mowing into things and voila - you got your autonomous lawnmower. For added benefit you can have all the computation done centrally and just give the lawnmower commands over let's say WiFi (and read data from the mikes).
The fact that sound moves so slow in the air will allow you to do something which you cannot do using radio - compare two signals.
You will not have any rodents, cats, bats or dogs in a mile radius either.
If you are that keen on doing it via radio use beacons at each corner and a rotating flat high gain antenna on top of the GAGA. Measure signal and triangualate. 4 WiFi access points, different SSIDs and a 20db directional antenna will do the trick - total BOM of under 300£.
In both cases the key is to stop occasionally, re-sync to a position and use inertial guidance in-between.
About as reliable as it is in the parking alarm sensor sat next to a chugging exhaust system. Or, in an old TV remote competing with the din coming out of the loudspeaker.
I can't see why it wouldn't work, myself. Couple of locator beacons on the lawn boundaries effectively broadcasting small signatures to say "I am boundary vertex number 0x0001", and some transducers (or a single rotating transducer pair) on the mower sending out a simple ping every 20ms or so and listening for the reply. The former lets the mower know where it is, the latter lets it avoid unexpected pets, toys and test engineers. So long as they are taller than the grass, of course.
Of course for extra propellorheadedness, make the whole system self-setting. Plonk the boundary beacons in, they talk to each other, find out their relative positions automatically, and feed that to the mower's onboard computer to tell it what part of the mowing boundary each beacon represents.
Then patent the lot and become the next Dyson. Muahahaha, etc.
> If you are that keen on doing it via radio use beacons at each corner and a rotating flat high gain antenna on top of the GAGA. Measure signal and triangualate. 4 WiFi access points, different SSIDs and a 20db directional antenna will do the trick - total BOM of under 300£.
I think sticking to radio with a rotating directional antenna could work well. Look at the Direction Finding sections of the ARRL Antenna Book and the ARRL Handbook for ideas. A rotating loop antenna (with an encoder on the spindle) could work quite well, then (with a loop) you just look for the nulls.
You will probably need 3 beacons, in an L. You then have three angles from "forward". You can use these angles to triangulate the position, plus you get the mower's orientation.
Alternatively you could obtain a goat.
Well, you might only need two if the beacons are just acting as locators, so long as the robot can't get to the other side of the line between them and confuse itself. I'm wondering how doable it is to have each beacon mark out the vertices of a polygon that describes the boundary of the lawn though. Stick a light on top and make them look like solar garden decorations.
If each beacon can talk to the others and they can triangulate their positions relative to each other, then they can chat with whatever system is aboard the lawnbot. Maybe even in real-time, so you can move the beacons around mid-mow and define areas where you can sit and enjoy your G&T while watching the mechanized underling beavering away in its half of the garden.
Like I said though, however you design the positioning system of this thing, I think it would be prudent to have a SONAR or RADAR system on the lawnbot itself for detecting obstacles. Nothing fancy, just a kit or COTS box that you point at something and it gives you a varying signal depending on whether it's pointing at anything and how close it is. Stick it on a stepper motor for 360 degree detection, and possibly hack it up with an MP3 player to provide random BOFHisms and threats toward whatever ankles, cats or plant pots get in the way.
Run a grid of wire under the lawn.
Have a low-powered transmitter on the bot.
You can then sense where the bot is by seeing which horizontal and which vertical wires have the strongest signal on them.
You could reverse it and have slightly different frequencies transmitted down each horizontal and vertical wire... might get a bit noisy though!
This problem has arisen in the hobby robot market many times try these solutions:
http://www.convict.lu/htm/rob/ir_us.htm
http://epia2011.appia.pt/LinkClick.aspx?fileticket=7DNWf7s86_s%3D&tabid=562
http://www.google.co.uk/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&ved=0CFgQFjAA&url=http%3A%2F%2Forbit.dtu.dk%2FgetResource%3FrecordId%3D221234%26objectId%3D1%26versionId%3D1&ei=1_5hT5fKCuqn0QWb9-ytCA&usg=AFQjCNEc5n3M-LQaKTsFYmXfvgihrpc6iw&sig2=h0xajW-1J-t_NYpg9MGZ6g
Two lasers mounted on a simple servo platform, one each in the bottom two corners of the garden.
A sensor on the bot.
First laser starts a sweep, bot reports when it sees a laser flash. This gives you tangent from one laser scanner.
Second laser does the same. This lets you triangulate the bot's position.
Put a rotating laser on the top of the lawnmower and (at least) 3 prisms in strategic places and then triangulate of them.
Otherwise, stick a total station in 'track' mode, whack a 360 degree prism on the top of the lawnmower and have the local station continually relay the position back to the lawnmower.
Of course, neither of these solutions are any use on undulating terrain.
This post has been deleted by its author
The film industry use high-contrast markers to identify parts of the image where they need to put some CGI in. See, for example, this article.
You could use slightly different markers in each corner - that way, the corner is uniquely tagged. Find the angle to each marker (camera on a stepper motor spindle, perhaps with a shaft encoder), and with knowledge of the garden dimensions, you can triangulate the current position.
There's a bit of coding to be done - but it looks like a fabulous job for a Raspberry Pi.
Vic.
I'd have thought that a wiimote style collection of IR beacons around the lawn and a decent camera on the robot would have done quite well.
How undulating is the lawn - From any point on my lawn I can see the two top corners of the house (i.e. ideal locations for beacons/reflectors)
For a LEGO robot I built. I never finished it, but it was due to time, and not a technical problem. I managed to get it to report its location fairly accurately, even!
First, it needs to map the yard - where everything is. Use proximity sensors on the two front corners to judge how close it comes to any given object. It drives around the outside, then fills in the middle.
Once it's mapped everything, you can download the map, and mark said map to tell it where it's allowed, and what boundaries exist that it missed (say, a gate), or that don't exist that it thought it saw (someone stood in front of it). With the right software, it can tell where it is by guessing its general location based on distance traveled, then match that location to a nearby landmark. Or, as long as it knows the outside boundary, just have it follow its own cut trail...
I would suggest an exterior brain, though; a computer that talks to it over WiFi, controlling its actions. That would give you a way to visualize its progress - and make it much smarter - without adding a lot of weight. That, and you can make it run on Perl scripts.
I agree with those other anonymous cowards. IR leds (powerfull ones if you want to work in the daylight) are the way to go. You could use a single "measuring stick". Personally I would see about a crude 360* camera or a panning camera installation and put a strobing led in each corner of the yard. Then you can use the strobe period to determine which led is which and use the relative angles to calculate your position.
Also, for image processing that measuring stick is by FAR not contrasting enough. If you want any chance of picking it out it should be dayglow orange or fluorescent green or something. Muddy brown just doesn't contrast enough to pick out of the background.
You could bury a grid of wires a couple of inches down in a couple of hours a tiny fraction of the time and cost of looking for a wireless solution. Either have the mower transmit a frequency into the wires and a base station report the position wirelessly to the mower or put an fsk id on the wires for the mower to id the wires as it crosses over them.
Instead of using visual recognition for distance, use triangulation. A couple of beacons with recognisable signals can give you directions. Measure the angles, calculate the location. For very rugged terrain, add more beacons. Navigators, pilots and automated systems use it a lot, and have for a long time.
I remember coming across an arduino based project that would seem to fit the bill nicely but can't for the life of me remember the search phrasing that it came up through but it went basically like this:
# Mobile unit sends out an ultrasonic ping
# arduino with three (or four) receivers around 'pitch' takes the differential timings of the signal to each receiver
# arduino calculates the location of mobile unit from the differential timings and locations of receivers
# arduin transmits coordinates back to mobile unit via Xbee
This method only requires accurate measurement of the locations of the receivers, not too difficult, I remember it saying they managed accuracy in the 10s of centimetres. Not too shabby!
How about some sort of buried cable. You could lay it out to suit a known path, Shove a current down it and you have a nice inductor under the ground throwing out a magnetic field. How you then track that field with the mower I have absolutely no idea. But, the idea is, you use that cable/field as a rail. The mower just starts at the beginning and ends, hopefully, at the end and runs into a brick wall or something to stop it.
Two Ultrasound transmitters that are separated and transmitting different but isochronous codes. (This can be done by "slaving one" off the other via IR link, so one transmitter is "dumb" just using an IR remote sensor driving the Ultrasonic transmitter directly).
A sensor on the robot can calculate the distance from each easily. If the robot is told what side of the line between the Transmitters it's on, it knows it's location. if it has a pair of separated sensors it can use the phase between them to calculate orientation.
Put the two Ultrasound TX at two corners of the boundary.
Check out the robot operating system at www.ros.org . Lots of people build lots of different SLAM solutions. One major problem is the correspondence problem, how to you know which spacial features are distinct. With active beacons this can be encoded. So have beacons chirping at different frequencies and then detect those frequencies by differencing an image at the same rates i.e. add the current image to the accumulator, then subtract it from the accumulator every X seconds. The accumulator with pick out pixels that vary at the same frequency, whereas other pixel values will cancel each other out.
Turn a camera into a omni direction camera by pointing it at a ball bearing. Bit of image warping and you can now get the *direction* of all distinct beacons. So now you can do "bearing only SLAM" (probably already in ROS I expect). A compass is also a useful addition for this kind of localization. Normal digital cameras = IR cameras after the removal of the IR filter.
Let us know if you need any help through Edinburgh Hacklab. We have alot of robotacists members (including me a.k.a. Larkworthy).
Have you seen the Nanotron NanoLoc devices? Uses chirp spread spectrum to get range information.
Of course two independant GPS receivers fix will wander relative to each other; they are free to use different satellites. Differential GPS involves generating corrections for individual satellites, not subtracting two GPS receiver fixes.
Maybe I'm missing something here, but why exactly are you trying to locate the mower externally? That's never going to be very reliable (as I predicted in the first round of consultation; my very own "I told you so" moment).
If you were keeping track of the displacement of the robot instead, turtle-like, it would be much easier. Distance is very easy, there are devices readily available. The direction would need to be worked out otherwise, but I guess that's not too hard. The angle of the wheels is not going to be accurate enough, but a fixed reference point should be good enough (IR light perhaps?).
If you know precisely how much you moved and in which direction, and you know where you started from, then you know where you are.
> What happens when a wheel slips on wet grass or mud?
Why would it matter? As I mentionned, measuring the angle of the wheels is not going to be precise enough even without slipping. Direction needs to be established thanks to a beacon of some sort. If you were referring to the odometer wheel, I think you'll find that odometer wheels are usually almost free-running, and thus unliquely to slip. If that worries you it is possible to use a studded wheel or even add metal crampons of some sort to make sure that the wheel won't slip. That is always going to be massively more reliable that any radiolocation, GPS, optical triangulation or any other method proposed to date. This is THE reliable technique. Plus, it's amazingly simple to implement.
A set of round posts with vertical bar codes. Or QR codes. Rotating sensor that knows the angle it's at with respect to mower front. Simple triangulation if 3 or more posts can be seen. And you'll also know which way the mower is pointing.
Codes could even contain info about the posts position. I.e. Sequence in perimiter, NE Corner, etc.
I think could be done as optical or radio location.
Basic principle - the beacon sends two signals. One, reference, omnidirectional, the other - modulated and directional, from a rotating emitter/antenna.
Measuring the difference between the reference and modulated signals will give your receiver's radial, i.e. the bearing from the beacon to the receiver. Using two beacons at known locations gives you the exact position of the receiver by simple calculation of intersection of two lines.
The receiver (GAGA) only needs to detect and decode the signal without having to identify precisely from which direction it is coming. So a simple radio aerial (if using radio beacons) or a fish-eye lens (if optical) will be enough.
If radio you can have a constant reference signal and AM or FM modulated rotating one, or a pulse reference with constant but very directional rotating one (so the radial angle will be calculated from the time between reception of the reference pulse and the peak of the rotating signal).
For optics - a strobe and rotating flashlight with some blinds for directionality and a colour filter for beacon identity could be an option.
"use a higher-power reader to read multiple tags" ... "we've not been able to find any hardware which will report the signal strength of an RFID tag"
Can you get a variable power reader? Sweep the power and note at what power level the various signals appear.
Or maybe have Kinect watch the mower moving?
I like the IR LED method mentioned here. By using multiple sets of LEDs in unique patterns the camera could measure the distance and compass bearing to whichever set it could see. Using the LEDs in pairs could also solve the problem of an LED going out. (Rare, but it's nice not to have to make repairs as often) Each pole would have an LED pattern set and a pair of reference LEDs below it so distance could be measured The Pattern Sets could be in an L shape with the reference LED a meter below. That way, the L Pattern could be turned 90 degrees for up to 4 unique patterns. Just remember to have a white L-Shape painted behind the LEDs on a dark background so the camera can easily see the shape night or day
Just come back from a prizegiving event for junior researchers where one of the winners was a kid who located his LEGO robot by using colour coding for the walls of its enclosure so the robot could see which way it was facing, and sonar for the distance to the wall. Worked reliably in a loud exhibition hall with changing lighting. The robot could maneuvre quite well inside the enclosure.
Replace the coloured walls with three or four differently coloured reflectors whose locations are known to the robot, and the sonar with Voland's ultrasound loudspeakers, and you're done.
A friend of mine solved this down to the sub 1M range for the seismic/surveying industry using inertial navigation married to GPS. Very simplistically, after the first 'good' GPS fix, all movement after that is relative, until the next 'good' fix etc.
Since you know 'exactly' where the garden shed is, as a starting point, GAGA could mow for x minutes (or to a pre-defined loaction) inertially, pause to acquire a 5, 7, 9, whatever satellite 'fix' , update its location if necessary and repeat as/if necessary until the job is done.
Size does matter, but since GAGA is a ride-on contraption, a designer backpack rig should not be a concern for the prototype.
I am not comfortable posting someone elses e-mail addy in a public forum, but if you wish to pursue, let me know and I will be happy to share with the SPB.
DP.
In case you missed it: This is a LAWNMOWER we are discussing. How is an ultrasonic transducer going to hear the echo of a tic-tic-tic over the din of a barely muffled single cylinder engine and the rattles generated by a slightly out-of-balance whirling blade?
Pulsed IR Beacons in the corners of the garden and a detector mounted on a goniometer FTW. You don't need to make the beacons transmit unique codes. You just need to strobe them so you can overdrive the power for extra brilliance at a low enough duty cycle to keep from smoking the junction. It also gives you noise rejection from steady-state sunlight glinting off of the prized glass garden orb.
The mower will start at a known location, and constantly update it's position from there. If you map the location of the beacons, then you should be able to quickly servo to one side of where you expect each beacon to be, Then do a slower scan to precisely locate that beacon. Then move on to the next one. Calculate your updated position while the sensor whirls back to start.
That's sorted, then. Oh, the weather was too nice for a coat, today. I'll just leave, then.
No difficulty in demodulation of PSK on a 40kHz narrowband carrier. Not a huge o/p from lawnmower at that frequency. 40KHz and 60kHz has been used in extremely noisy environments.
The data rate doesn't need to be high which lowers the SNR needed.
This post has been deleted by its author
You put a motorized camera on top of the thing, to take panoramic photos of its surroundings.
Using these images, a suitable application scans the images and finds 'important' points. After taking one of these panoramic views, the robot slowly displaces itself a given distance and repeats the process, until it has gathered enough data to triangulate the entire garden. From this data the app obtains a 'mathematical model' of your garden, that it can use to choose its path .
If you want to define 'boundaries', so your Gaga-bot doesn't try to mow the swimming pool, you use a 'laser wall', something you can cheaply build yourselves with a laser pointer and a suitably shaped reflector. Once the bot has 'learned' it's surroundings, the user can save and give a name to the 'map' generated. After this, the user can remove the 'laser walls', as they aren't need now that Gaga has the terrain map.
In normal operation, the Gaga takes more photos -not complete panoramics- to know its exact position and define an optimal path using the map model created before.
I admit that the programs to perform those tasks must be really complicated, but several systems for obtaining 3D data from overlapping 2D photos have already been developed, and this is very close to what would be needed here. I'm sure that if ElReg sponsored an open source project for this, they would get top talent to do the work. And lots of fun.
Another drawback is that you'd probably need a PC connected -WiFi?- to the Gaga-bot to perform these tasks. Of course that's not a big issue, as any one nerdy enough to build one of these, and use it, probably has more computers at home than they have shoes.
Yes, I mean all those guys able to lay relaxedly in a hammock enjoying a pint of beer or three, while Gaga suffers some software bug, escapes the boundaries of his internal map and brings indescribable havoc upon the neighborhood.
"You put a motorized camera on top of the thing, to take panoramic photos of its surroundings."
That'll fail. Because of at least 2 reason: no motorised camera is going to report it's facing direction accurately enough for that use. And building the panorama while the mower is moving would render the resulting panorama completely useless anyway.
You need a 360 degrees camera.
The third reason why it will probably fail is that with vibrations, light variations etc you will probably never be able to reliably recognise the features. Certainly not reliably enough to do any kind of triangulation (that's true for most of the proposed optical triangulation methods, btw).
I stand by my guns; the only thing I'm still working on is the precise calculation of the curves (as in my proposed method, errors are cumulative, that's potentially a BIG deal).
my mum has a gardener go round every few weeks to mow the lawn, he charge £10 an hour and it takes him a couple of hours... that's £360 a year
is it really worth all the faff and hundreds of pounds of kit to avoid paying less than £400 a year to help out a local small business?
I'm pretty sure whatever kit you end up with it will need fixing/replacing every few years anyway
Although I have barely even dipped my toe into this field at the moment, both the 3d scanning of objects, and the 3d printing of objects are projects that I am considering getting into currently.
There are several projects that are using the kinect camera for a basic architectural survey, it is able to pick up features that experts can determine need more investigation. The more passes the camera does, the more detail gets overlayed.
You would need to rotate the camera through a 360 degree rotation at a known fixed speed, and then compare the created image against a predefined detailed map
Looking also at the requirements of double or triple passes and number of blades, you could have the motor providing the movement toggling between moving, or scanning its location.
If it is not powering the wheels for movement, the power goes into the turntable / stepper motor that turns the camera and processing unit through 360 degrees. This would give you more time in each spot to get the quality cut you are looking for.
By separating out the task (cutting the grass) from the scanning, location and movement, you end up with two different projects that may be useful in their own right.
Are there any pictures of the actual lawn and it's boundaries online? so we can gauge the size of the task in hand?
http://www.culture24.org.uk/science%20%26%20nature/technology/art369765
At robotville there was a robot that used a Kinnect and lasers to map out where all the walls were in the room so that they could know where they were relative to their starting position. Don't know if that is overkill though.
I worry that you may be making this too complicated; surely the easiest way would be to surround the lawn with some kind of barrier (e.g. a buried wire carrying a low frequency signal). While barrier mapping is good if you can see a barrier sticking up (e.g. a wall), it would not prevent the lawnmower from falling in a pond or swimming pool (of course, other sensors could be used to detect such things, but they add more complexity). I'm guessing that you intend to program a map of the lawn manually, so that the lawnmower knows where your garden stops and your neighbour's car begins; if so, you had better get your measurements spot on!
Burying a cable, of course, would eliminate the need for mapping beyond a simple inertial model; the lawnmower could roam around the lawn like a demented Roomba, and you would never have to worry about it escaping. Unless the low frequency power went off, of course.
I did have a crazy idea about RuBee style inductance transmitters; I'm no expert, but I understand that if the ariel size is a tiny fraction of the wavelength, the magnetic field dominates the electric field. As the magnetic field falls off much more quickly than the electric field, you could triangulate from magnetic field strength with greater accuracy than from the electric field strength, while still avoiding interference. All you would need would be a few short ariels / inductance coils transmitting different VLF frequencies (with kilometer-scale wavelengths). (These low frequencies propagate without worrying about water, metal, humans etc., which would solve one of your WiFi problems.) The lawnmower could then triangulate its position with some simple analogue electronics.
No idea if this would work, but it would be reasonably easy to try, and would work a lot better than using RFID tags which are *designed* to be unreadable more than a few centimetres away.
Gardens aren't just empty space, they have things like swing sets, paddling pools, trees and sheds in them. Obstructions could cause problems for beacon-based optical methods.
The guy who suggested dead reckoning using wheel motion got downvoted, but it is a serious technique. Check out this article where 0.5% accuracy was achieved:
http://geology.heroy.smu.edu/~dpa-www/robo/Encoder/imu_odo/
Recalibration could be done at close range to a fixed beacon, which would be simpler and cheaper than long-range positioning.
In the past ultrasound has solved this problem, at least indoors.
Your problem with GPS is the speed of light, but you don't need a GLOBAL positioning system, you need a local one. Ultrasonic transducers of the kind often used in car alarms are easy to get hold of, they have a peak efficiency somewhere around 40khz but you need two or three distinct beacons. Just below, at and above that point. More beacons make obtaining a fix easier.
Synchronised pulses from multiple beacons reach the robot at different times, or you can send pulses from the robot to be located by multiple receivers, whatever, you get the picture.
I'm all for putting a beacon on the bot and a camera on the fence. put it in the corner and it can do an L-shaped garden, although it seems reasonable not to support crazy shapes - most gardens are more or less rectangular. The bot would only have to flash a signal every so often, so it'd be very low power, and the camera could be hooked up to the mains. Super cheap and ticks the boxes.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
