Subject:
|
Re: Line following algorithms
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Mon, 25 Nov 2002 20:17:07 GMT
|
Viewed:
|
946 times
|
| |
|
|
In lugnet.robotics, Steve Hassenplug writes:
> I keep wanting to make up something like this. But I haven't, yet. Here's
> something to get you started:
>
> http://www.geocities.com/laosoh/robots/linefollow.htm
>
> Steve
Thanks Steve,
I saw these site before, it´s a nice job but the described algorithm is
based more on "edge following" than on "line following". For this kind of
algorithm the best solution would be a grayscale track such as the "Nascar
racers" proposals on another discussion in this newsgroup. I tried it some
weeks ago and it really works well in this kind of track, but the track
building is not a simple task with home made material such as A4 printed papers.
I would really know the state of art in line following algorithms, his pros
and cons, and perhaps made some enhancements on some of them.
I´m a C programmer for 15 years in industrial automation area,
unfortunatelly I discovered Mindstorms just a month before, but i became an
AFOL immediatly and bought 2 RIS, 1 RDS and a DDK since them.
I conducted some tests with line following and bar code reading (all of them
with 1 single sensor) mainly with Mario Ferrari´s Quite C environment (I
have not currently explored LegOS stuff).
I tested the canonical algorithm as described at MIT site, but they behaves
poorly on 90º turns.
The best working algorithm I found until now is the simplest one:
1) When light sensor on track goes ahead.
2) When not on track, seek track in right or left (randomly) direction for
some time.
3) If found, go to step 1.
4) If not doubles seeking time, reverse seeking direction and go to step 2.
It works, all tested combinations of turns and straights worked, but the
drawback is that it stops too often to "find the track". I´m currently
working on a manner to better predict wich side would the best for seeking
first based on some statistical way (better than ramdomly).
I will also better the robot´s based on the derivate of the changing light.
If the light changes too fast, the robot must stop immediatly and begin a
"find a track" procedure.
If the change is slow he can search for the track without completely
breaking, just slowing down one motor and see if the light becomes dark
again, otherwise, the seeking side must be reversed.
Any comments or suggestion will be wellcome.
Paulo
P.S.: Sorry for poorly written english. It´s not my first language.
|
|
Message has 2 Replies:
 | | Re: Line following algorithms
|
| (...) Dave Baum's Definitive Guide gives a good treatment for the case of a single light sensor. However, to follow a line as opposed to "edge following" with elegance and ability to negotiate sharp 90-degree bends requires two light sensors as was (...) (22 years ago, 26-Nov-02, to lugnet.robotics)
| | | Re: Line following algorithms
|
| (...) Have a variable remembering if the 'bot found the line turning left or right and try that direction first instead of random. This way the bot will go much fast through long turns but will always seek wrong direction for the first correction of (...) (22 years ago, 27-Nov-02, to lugnet.robotics)
|
Message is in Reply To:
14 Messages in This Thread:
  
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
This Message and its Replies on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
