robot navigation

[Paul Obrien]

AllanBertelsen, sounds like yo're after a dead reckoning system, which could include accelerometers and/or fibreoptic gyros with some reference checkpoints

 

[Paul Obrien]

Ok, after checking out some fibreoptic gyros maybe the aren't the way to go as the MiniFOG weighs 35Kg. Way not try encoded sensors on the wheels. Optic disc won't be feasible as it's on a mower with grass and dust being blown everywhere so why not try a ring with magnets and Hall effect sensors.

 

[dknguyen]

Just go for an piezo gyro from Analog Devices or a ring gyro from Silicon Sensing. Pretty much no hobbiest (I don't want to say for certain since you can't know if black swans exist until you see one) uses optic gyros...way too expensive.

And some MEMS accelerometers from Memsic or Analog Devices.

Wheel odometry will get way off very fast because wheels slip...especially on bumpy, uneven surfaces like grass. PLus I just think it's easier to toss an accelerometer and gyro IC onto your circuit board rather than having to deal with all the mechanical aspects and mating of an encoder.

 

[bananasiong]

bananasiong, does it actually mow anything?

Hi,
it doesnt mow anything...but it mows the grass.. LOL. u can't see from the clip. someone was asking about the navigation, that's y i post this up. and.. actually the mowing system is not working quite well, coz i'm using plastic blades.

 

[Paul Obrien]

What never seen a black swan!

 

[AllanBertelsen]

Just go for an piezo gyro from Analog Devices or a ring gyro from Silicon Sensing. Pretty much no hobbiest (I don't want to say for certain since you can't know if black swans exist until you see one) uses optic gyros...way too expensive.

And some MEMS accelerometers from Memsic or Analog Devices.

Wheel odometry will get way off very fast because wheels slip...especially on bumpy, uneven surfaces like grass. PLus I just think it's easier to toss an accelerometer and gyro IC onto your circuit board rather than having to deal with all the mechanical aspects and mating of an encoder.

Wheel odometry, accelerometers, gyros and compass is going to be included one by one. But there will be drifting for all theese techniques. I will also use ultrasound proximity sensors to avoid killing a cat or runing over my kits feets.

The thing I need is a way to precisely correct any drift errors. For example when the mower stops at the end of a run before turning.

dknguyen mention occasional checkpoints. That could very well be a good approach. Would IR be disturbed by sunlight? Coils or pices of iron bearied at strategic places or other tings that easily can be detected.

The 80m radio beacon is an interesting thing to examine. I just don't got any experience to make any assumptions about precision. And I can't find anyting about it at the Internet. So I hoped that some expirienced radio amature could comment this idea.

 

[dknguyen]

I believe the very little solar IR gets through the atmosphere. If I remember right, the sky looks black in an infrared photograph photo.

You can help correct for bias drift by having the "zero" output voltage recalibrated everytime the robot comes to a stop (ie. the motors have been deactivated with brakes on for more than 1 second). This will fight the drift involved in the sensors' bias. You might actually want to program your robot to come to a stop for at least 1sec every time it stops so you can do this calibration more often. This should be effective in fighting against drift due to temperatures and other such things (it won't correct for the sensor's scale factor though)

If you used IR/sonar beacons then you could be more accurate for longer periods of time, with only drift-vulerable sensors filling in for the beacons when you lose contact. Basically with this method you are using localized "IR/sonar GPS" with the inertial sensors filling in for when you lose contact until you can restablish contact. Just to reclarify, the IR/sonar method sends an IR signal and a sonar pulse at the SAME time. Because light and sound travel at different speeds they will reach the robot at different times. You can use this time difference to calculate how far your robot is away from that beacon, enabling you to draw a circle around the beacon indicating possible positions of your robot. If you have 3 beacons (most effectively spread out to surround your robot from all directions) then you have 3 circles which will only intersect at one point. This is where your robot is. (Of course, reality will not make a perfect intersection, you will have to use an algorithm to narrow down where you might be in the "gray" area. Read up on GPS and see how they do it- to put it simply, they adjust the all the circles in EQUAL amounts until an intersection is made. You could also find which point in the gray area has is the same distance to each circle's edge.

If you want REAL check points (as opposed to beacon-positioning with inertial dead-reckoning during times of lost contact) then use RFID tags or something since their range is much smaller than IR and does not suffer from light of sight. If you sense one, you know where you are with much greater certainty. using just IR for a beacon, may not be a good idea since it's too easily visible from far away.

RFID tags are way simpler to use. YOu don't need to build a minimum of 3 IR/sonar beacons. You don't need position the beacons around your yard every time. You can just hang a few RFID in an RF transparent weatherproof plastic box in some trees or something (and make sure your robot knows where they are supposed to be). And you if you use the passive kind you need no power supplies. The price you pay is that if probability plays against you and the robot does not pass by a RFID tag for a long time, it's position errors wll continue to accumulate. But as soon as you hit a tag, bam, your error is reset.

 

[Nigel Goodwin]

There's loads of IR outside on a sunny day, which is why IR remotes don't work well, as the receiver gets swamped.

I built an IR remote tester years back, and even inside during the summer you had to hide the tester in the shadows due to IR pickup from the sun - it was OK though for probably 8-9 moths of the year!.

Built a new one since, this one uses the 38KHz receiver IC from a TV

 

[AllanBertelsen]

I have been investigation RFID. It might be a good way to correct positioning errors. Only thing is that they only operate at very close range. If the robot has to return to a tag each time it need correct its position, it will be inefficient. And when it returns to where it is supposed to continue its work, it will be wrong positioned again.
Look at this link:
**broken link removed**
This is for VHF fox-hunting. Very accurate. 80m is another wave-length. But the principle could be used on 80m receivers I mentions in earlier posts. This principle eliminates the need for a rotating antenna.

 

[dknguyen]

Don't have it return to checkpoints. Use a few of them evenly scattered, so that whenever the probability arises that the robot passes by one, it will know it's current position. Let the robot move around however it wants to move around. Eventually it will run by a checkpoint and have it's position info corrected.

 

[billericayboy]

In your first post you say that a Buried Wire containment system or fence is not practical but there are several commercial robot mowers on the market using this system very sucsessfuly.

I am looking at building my own system and have found the following site www.mowbot.org which details a guy's quest in making a robot mower. i have yet to build the fence and sensors (for two reasons 1. i am ok with a bobot with bumb sensors but am a total novice at electronics and 2. i do not know how to measure and calibrate a fence/sensor to work together).

If anyone has built a buried wire fence then i would love to hear from them, or even any suggestions on the best way to build my own system.

Below are links to the transmitter & receiver i have found.


http://www.betuwe.net/~mowbot/UDO5transSch.PDF
http://www.betuwe.net/~mowbot/UDO5recvRcv.PDF

Thanks

 

[AllanBertelsen]

Thank you for the link. It's very inspiring. Problem is that my lawn is 5000 m2. It's subdivided by beds and borders. The system has to be bigger and cleverer.

 

[billericayboy]

Buried Wire Fence / Sensor

My mother is using a MOWBOT www.mowbot.co.uk and i had to install a wire around the perimiter and then ran the cable from the perimiter to the fish pond, around it and then back to the perimiter down the same slot in the lawn as the wire coming in to the pond. one wire cancels out the other, so you effectivly just have a mini-boundry around the pond or any other obstical. The Mowbot runs on a sinusoidal wave as well but i cant get any specification form the UK dealer and i dont have the equipment to measure what is being sent down the cable.

Any suggestions, i am waiting for a friend to lend me a Frequency Counter to see if i can pick up the frequency it uses and then the task of making my own transmitter to build a loop of my own. Any suggestions would be of great assistance.

 

[ikalogic]

hello everybody

this article on my website tries to deal with this problem.

**broken link removed**

It's a dead reckoning robot.

hope you'll find it useful

 

[AllanBertelsen]

Good work Ibrahim Kamal. You would be able to implement real PID control with the robot you got there.

 

[ikalogic]

Good work Ibrahim Kamal. You would be able to implement real PID control with the robot you got there.

i am not really sure... the feedback is too poor to implement PID.

i am actually working on another robot, where the mechanics are all made by a team of 6 robotics hobyist (including me) where high quality feedback is implemented on most of the motors, and in that robot, implementing PID over many parts of the control system was really very interresting...

 

[bobledoux]

Dead reckoning comes from aviation. The airplane flies to a distant point, determined by a radio beacon. The intermediate path is not important. If the airplane is blown off course by wind the path to the beacon changes. The airplane still arrives at its destination.

In lawn mowing the path is important. Each path must overlap the previous path. That requires monitoring position and heading, very precisely. Error correction requires rapid return to intended path or spots of lawn are missed.

My project developed from an approach performed by a neighbor. He had a small tree in the middle of the yard. A rope was attached from the tree to a self driving mower. As the mower circled the lawn the rope wrapped around the tree, bringing the mower path closer to the tree. When the tree got bigger he needed a wider mower.

I attempted to generate a similar method using a central beacon. It transmitted a radio and ultrasonic beacon. The time delay between the robot reception of the two signals indicated distance from the beacon. The robot steered circular paths of a given radius, based on the signal and its changing values. It sounded good in theory, but the ultrasonic signal had lots of false receptions.

One possibility is to bury a wire that represents a path for the robot, like a line following routine. Since a precision path is required this might be the best approach.

 

[billericayboy]

Could you not use the wire to make a perimeter and the robot cuts up to the wire then turns around and wanders off. You may cut the same bit of the lawn several times but it will all get cut and saves laying a very long wire up and down the lawn.

The robpt can also follow the wire like the Robomow, Mowbot etc. when they go back to their charging bases.

 

[bobledoux]

I think we are trying to move beyond the random pattern technique. A more sophisticated method has broad applications, for example buried mine location.

Imagine your random movement robot mine identifier has searched an area for two days. Computer models suggest 98% of the mines will be located after this period. Is that enough confidence for you to walk in and dig up the mines that have been located? Where are the other 2%?

The random pattern is effective in low effort applications. It is not efficient in that areas are covered multiple times and small areas are missed. For lawn mowing, re-cutting an area is no big deal. But if a tractor is pulling a planter across a large field over drilling seed stock is a big deal.