Subject:
|
Re: Positioning
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Thu, 27 Jan 2000 19:47:13 GMT
|
Viewed:
|
922 times
|
| |
|
|
Hi Marco,
Building the robot is definitly fun. But I was really suprised when I built
two robots from the same design, ran the same "GO" program in them and they
moved differently! 8^o
I'm glad to find another ant lover here! I try to check all my designs after
ant abilities. When I built RoboCall I was worried that it only supported 15
different message types until I read that the average species has only 12 ways
of communicating. I use the book "The Ants" by Bert Holldobler and Edward O
Wilson to check my designs and mine for ideas.
My position on the mapping stuff is that there are many ways to do a map and a
set of instructions to follow the map. Each way has strengths and weakness and
you should pick the method that best matches the problem you're facing. In my
mind, here's how the Timing, Rotation, and Landmark methods stack up:
TIMING
ADVANTAGES: 1) Times are consistant across different robots. 2) Simple.
Tracking time is simple math.
DISADVANTAGES: 1) Can be thrown off by wheel slippage. 2) Changing wheel sizes
or gear trains completely throws off your timing code. 3) The path followed by
the robot can never change.
WHEN TO USE: When you're following a precise route that will never change on a
smooth surface with a single robot.
ROTATION
ADVANTAGES: 1) Simple. Tracking rotations is simple math.
DISADVANTAGES: 1) Inconsistant. The longer you use a rotation sensor, the
greater the error in its measurement becomes. 2) Can be thrown off by wheel
slippage. 3) Changing wheel sizes or gear trains completely throws off your
rotation code. 4) The path followed by the robot can never change. 5) You have
to buy a rotation sensor and learn how to use it.
WHEN TO USE: Honestly, the only reason I see for using this method instead of
Timing is you want to play with a rotation sensor. If there is a situation
where the routes are precise but for some reason Timing doesn't work, you
could try this method.
LANDMARK
ADVANTAGES: 1) Immune to wheel slippage. 2) Immune to changing wheel sizes or
gear trains. 3) Can follow a path even when the distances between landmarks
change.
DISADVANTAGES: 1) Other than finding colored strips on the floor, finding
landmarks is HARD. 2) If the robot misses a landmark, it just keeps going, and
going, and going...
WHEN TO USE: When you're following a route where the distances between
landmarks can change. When you're running different types of robots, with
different gear trains, along the same path.
No method is perfect, and the programmer will have to write code to handle
errors when (not if) they arise. In the case of a rotation sensor, it's
cummulative error can be greatly reduced by attaching it to a worm gear. In
the case of landmarks, a second map can be used to let the robot figure out
when it's off track. The second map may be based on timing, as in "You've gone
a minute without finding a landmark, stop and go back to the last landmark and
start again, or on other landmarks, as in when you see a black line, you're at
the edge of the arena, stop and go back to the last landmark and start again.
That's my view of these three methods. I'm sure others will have different
views.
David Leeper (looks forward to hearing about your vision system)
In lugnet.robotics, "Marco C." <marco@soporcel.pt> writes:
> At 08:54 27-01-2000 GMT, David Leeper wrote:
> > 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.
>
> That's one of the reasons why a programmer goes to all the trouble (and
> fun?) of building a physical model of a bot instead of dealing with
> over-acurate 3D CG simulations.
>
> Ok, now for the Positioning system debate. Just one word: *Ants*
> They're blind. The advantage they have is chemical sensors (smell).
> I think they use a mix of "(chemical) line" following and landmark system
> (or line node recognition)
>
> My robotic vision project will, hopefully, evolve to something that can be
> used as a landmark/graphic pattern recogniser to be used in the bot
> positioning area.
>
> :)
>
>
> PS: MyProjectsUpdate:
> My Robotic Vision project (using http://www.ccs.neu.edu/home/psksvp dll's)
> and my Sanscript & Spirit.OCX project (http://www.trulyvisual.com) have all
> stalled temporarily. I was expecting this already, 'cause this time of the
> year my workload doubles, so no spare time :(
> ..best guess is March, to continue with both projects.
>
> ____________________
> Marco C. aka McViper
|
|
Message is in Reply To:
 | | Re: Positioning
|
| (...) come (...) That's one of the reasons why a programmer goes to all the trouble (and fun?) of building a physical model of a bot instead of dealing with over-acurate 3D CG simulations. Ok, now for the Positioning system debate. Just one word: (...) (24 years ago, 27-Jan-00, to lugnet.robotics)
|
28 Messages in This Thread:
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
