Subject:
|
A robot who knows his position (fwd)
|
Newsgroups:
|
lugnet.robotics
|
Date:
|
Wed, 28 Apr 1999 01:08:14 GMT
|
Original-From:
|
Jim Choate <ravage@EINSTEIN.ssz.com*antispam*>
|
Viewed:
|
897 times
|
| |
|
|
----- Forwarded message from Mario Ferrari -----
From: "Mario Ferrari" <mario.ferrari@edis.it>
Date: Tue, 27 Apr 1999 22:12:25 GMT
I had some success in my first experiment with odometry. My goal was to build
and program a bot that at every moment knows where he is.
The equations to deduce the bot position from wheel movements are pretty easy
to use and understand, but I had to face the problem that they require floating
The representation of the space inside the bot is a simple Cartesian plane,
where the origin is the starting point. Distances are expressed in 10000th of
millimetre and angles in 10000th of radian. The bot is obviously not precise up
to this point, but just to keep computation approximations as low as possible.
----- End of forwarded message from Mario Ferrari -----
Jesus H. Christ, talk about over-kill. If all you're looking for is position
sensing then the absolute highest resolution that is required is the
dimension of the robot. Convert this to a integer multiple of wheel diameter
and you can loose the floating point math, it's all integers; 1 body, 2
body, 3 body, etc. You'll still need floating point to do the vector algebra
but as you've discovered, tables are the way to do that.
As to the smallest usable dimension, the question there is considering the
smallest resolution in your A/D for wheel rotation and then using it as your
base ruler for fine positioning, you can't get any finer than this.
What you want is two levels of precision. The first used for gross
positioning and obstacle mapping. The second fine-scale used for close in
maneuvering and goal completion (ie dropping a can in a bucket).
If you're talking about a free-roaming bot that could live in the outdoors
(ie unlimited world dimensions) then you want to get rid of the
dead-reckoning vector math and go to some sort of inertial guidance. Wheel
slippage, precision errors, motor fluctuations, battery drain, etc. make it
less than effective. The errors pile up so fast the bot gets lost pretty
quickly.
Current small-scale, single-unit purchase, accelerometers are going for about
$300 now. Real popular with the high-performance and experimental (go bird, go)
rocketry crowds.
This months EDN has a add in the rear for new products for just such a
device.
____________________________________________________________________
Three step plan: 1. Take over world. 2. Get lot's of cookies.
3. Eat the cookies.
Anonymous
The Armadillo Group ,::////;::-. James Choate
Austin, Tx /:'///// ``::>/|/ ravage@ssz.com
www.ssz.com .', |||| `/( e\ 512-451-7087
-====~~mm-'`-```-mm --'-
--------------------------------------------------------------------
--
Did you check the web site first?: http://www.crynwr.com/lego-robotics
|
|
Message has 2 Replies:
 | | Re: A robot who knows his position (fwd)
|
| (...) up (...) possible. (...) That's true and clear. I'm not actually interested in such a precision, but as you know very well the problem with odometry is that errors accumulates very fast so I used four decimal digits in any computation to keep (...) (25 years ago, 28-Apr-99, to lugnet.robotics)
| | | Re: A robot who knows his position (fwd)
|
| (...) [ talk of Dead Reckoning ] (...) That's a useful idea; I'll take note of that. (...) I've been doing some robot building aimed at dead reckoning and have found that even the home or flat environment is enough to f*&k things up; the bumpy (...) (25 years ago, 28-Apr-99, to lugnet.robotics)
|
14 Messages in This Thread:
 
 
 
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
