Subject:
|
Re: Positioning
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Mon, 31 Jan 2000 03:47:06 GMT
|
Viewed:
|
2683 times
|
| |
|
|
Hi Mauro,
I understand now, and you're right, it is a concern. I think the way to deal
with it is to make landmarks big and hard to miss. This is the way landmarks
work in real like. They are big. In this case, I'd say this means using a wide
strip of paper. There would still be error in turning and moving, but that's
due to the nature of robots. I would think the error should be small enough
that the landmarks system would work for several turns, enough to perform most
tasks.
David Leeper (doesn't mind "cheating" to get the result he wants)
In lugnet.robotics, Mauro Vianna writes:
> Well, I can tell the robot to turn 45 degrees using timing but it's not
> precise enough. It would be something about 43-47 degrees. I could use a
> rotation sensor (but I don't have one) to increase accuracy but there will
> always be a smal error. After navigating through some landmarks there will
> be a cumulative error that's not small anymore.
> Things get worst when you use mixed terrain since the timing necessary to
> turn 45 degrees is different over a rug than it is over a plain wood floor.
> Sometimes there is a small direction change when the robot is moving
> straight and changes from plain floor to a rug. Add this to errors related
> to 2 motors robots (left-right) or steering robots and you have a lot of
> small errors.
>
> Resuming, There will always be a cumulative error, which is agravated in
> mixed terrains. I believe we will eventually need a course correction to
> "reset" this error. You can think about this as a calibration. I described
> one but didn't test yet.
>
> Mauro Vianna (trying to express himself better)
>
>
>
> "David Leeper" <david_leeper@hotmail.com> wrote in message
> news:Fp4499.LAB@lugnet.com...
> > Hi Mauro,
> >
> > Maybe I'm misunderstanding, but it seems to me that if you can't tell your
> > robot how to turn correctly, no mapping system would work.
> >
> > The system I used contained a simple TURN LEFT, TURN RIGHT method. Each • turn
> > is 45 degrees. I didn't really have any problems with this.
> >
> > David Leeper (may have missed your point)
> >
> > In lugnet.robotics, Mauro Vianna writes:
> > > I have been watching this discussion with great interest. I was looking • for
> > > a solution for this problem since I wanted to program a robot to find his
> > > way inside my apartment.
> > > The problem I found in the landmark system was that you don't have a • precise
> > > way to tell the robot how much to turn when it find it. Since this error • is
> > > cumulative it will evetually lose the landmarks. It will find the first, • the
> > > second but will eventually miss. It is especially important when the the
> > > robot moves through different knids of terrain (carpet, plain).
> > >
> > > Did anybody try any kind of course correction? I thought about one but
> > > didn't try it yet. The landmarks could have short lines of different • color
> > > extending from it in the direction of the next landmark. This way an
> > > algoritm similar to the line folowing could be used to set the robot
> > > direction.
> > >
> > > Ok, ok, the problem of landmarks on the living room vs. wifes persists... • :)
> > >
> > > Mauro Vianna
> > >
> > > "David Leeper" <david_leeper@hotmail.com> wrote in message
> > > news:Fp0v94.1r1@lugnet.com...
> > > > Hi Pete,
> > > >
> > > > I agree that the short sight is a problem. In my tests, I used wide • strips.
> > > > Three times as wide as the RCX. And they were no more than a foot • apart.
> > > So
> > > > missing them would have been difficult.
> > > >
> > > > In real life, you could use similar tricks. For example, suppose you • wanted
> > > > your robot to travel from the bedroom, into the living room, go over to • the
> > > > corner and turn on the stereo. You could put a red rug in front of the • stereo.
> > > > That's your landmark. People visiting you wouldn't think you're strange • (like
> > > > they would if you layed colored strips of paper all over your house) • and
> > > the
> > > > robot would have a (say) 1 foot by 3 foot target to hit.
> > > >
> > > > You could also combine the mapping methods. Since the relationship of • your
> > > > bedroom to your living room is unlikely to change, you could use a • timing
> > > map
> > > > to navigate that. Then use a landmark map to zero in on the stereo. If • your
> > > > girlfriend moves the stereo a few feet to make room for her plants, • your
> > > robot
> > > > still finds the stereo.
> > > >
> > > > And, of course, I'm hoping sonar will increase the range that landmarks • can be
> > > > spotted. I'm not expecting it to be easy, and I don't even know what a
> > > > landmark would look like to sonar, but I'll find out more this summer.
> > > >
> > > > And in talking about all this, it's starting to dawn on me that error • handling
> > > > while traveling a map is relatively unexplored territory.
> > > >
> > > > David Leeper (is still learning)
> > > >
> > > >
> > > > In lugnet.robotics, Pete Hardie <pete.hardie@dvsg.sciatl.com> writes:
> > > > > David Leeper wrote:
> > > > > >
> > > > > > Hi Pete,
> > > > > >
> > > > > > I was looking back over you post and realized I didn't answer a • couple
> > > of
> > > > your
> > > > > > questions.
> > > > > >
> > > > > > > what are the steps leading up to "...when you get to the X..."?
> > > > > >
> > > > > > In the program I wrote there are only a few commands. Go forward. • Turn
> > > left
> > > > at
> > > > > > landmark X. Turn right at landmark X.
> > > > > >
> > > > > > > And how does
> > > > > > > a bot determine that it has not found a landmark and needs to ask • more
> > > > > > > directions.
> > > > > >
> > > > > > To be honest, I ignored the problem. Although I knew it could happen • in
> > > > > > theory, in practice it never did, though I admit it was a limited • set
> > > of
> > > > > > trials in a static environment. Offhand, I can think of a couple of • possible
> > > > > > solutions. 1) Just stop moving after a given time limit or wheel • rotation
> > > > > > count. 2) Have a set of landmarks that tell you when your lost, as • in
> > > "If
> > > > you
> > > > > > see a McDonalds on your left, you've gone too far." Really, both • answers are
> > > > > > the same. Have a mapping system to tell you when you're not • correctly
> > > > > > following your mapping system.
> > > > > >
> > > > > > Again, I think the major flaw in all mapping systems that I've seen • is
> > > > > > the "GO" command. Moving a robot is not acurate and the programmer • has
> > > to
> > > > come
> > > > > > up with ways to deal with the inacuracy, whether its a timer system, • a
> > > > > > rotation system, or landmark system.
> > > > >
> > > > > The main problem I'm seeing is that the legobot has very short sight, • and
> > > so
> > > > it
> > > > > can
> > > > > only travel a certain distance before it is nearly sure to miss a • landmark,
> > > > and
> > > > > then
> > > > > be lost.
> > > > >
> > > > > Also, in the Real World (tm), creatures with the limited senses of • legobots
> > > > tend
> > > > > to
> > > > > use trail-following methods (ants and scent trails, etc) as the main • path
> > > > > tracking,
> > > > > and perhaps landmarks to ID branching points. Creatures with • long-range
> > > > senses
> > > > > use distance measures (bees and the honey-dance)
> > > > >
> > > > > --
> > > > > Pete Hardie | Goalie, DVSG Dart Team
> > > > > Scientific Atlanta |
> > > > > Digital Video Services Group |
|
|
Message is in Reply To:
 | | Re: Positioning
|
| Well, I can tell the robot to turn 45 degrees using timing but it's not precise enough. It would be something about 43-47 degrees. I could use a rotation sensor (but I don't have one) to increase accuracy but there will always be a smal error. After (...) (25 years ago, 30-Jan-00, to lugnet.robotics)
|
28 Messages in This Thread:
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
