Correcting a Rate Gyro Issues?

Hi all, I've recently included a gyro on my robot to enable me to determine direction, as working out from wheel measurements got way off very
quickly. It's make a really good improvement, however it still ends up in the order of 30/40deg's out after about 10mins of running about...can get wore depending on how much turning the robot does in that time.
I've had a look on the net for solutions to this, and I found a good one on the Seattle robotics page, where a digital compass is used.
I'd ideally like to use some other kind of sensor which would be affected by factor's external to the bot if possible (i.e. earths magnetic field) as I want to mount it inside the bot where there are all sorts of electro-magnetic transmissions - RF/motors etc.
Would an accelerometer help? Any one tried this?
Thanks and Regards
Mark
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

A thought...
Wheels can slip throwing off your measurements through repeated turns over time. You need to measure the movement of the robot with respect to the ground in a manner which is absolute, rather than the rubber on floor method. Consider using the data stream from an optical mouse. Think of the floor as an encoder pattern. You could attach a lens to bring the floor into focus from whatever height the mouse has to be mounted at. Using this approach you could even measure and publish the amount of tire slippage you are seeing.
Of course, this requires the existence of serial optical mice. Is there such a thing? Just a minute... (google) ...
Hmmm... this is interesting, but not what I'm looking for: http://www.home.agilent.com/USeng/nav/-536893734.536883737/pd.html
Ok... There appear to be PS/2 optical mice available, no serial, and USB. Serial appears to be out. USB is out because the USB host thing is near impossible.
PS/2 mouse information: http://www.computer-engineering.org/ps2protocol / http://www.computer-engineering.org/ps2mouse /
Let me know if you try this.
Brent S.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
Brent, Not really an option am afraid.
Looking for a more accurate approach (Optical mice can "slip" as well). I thought I'd seen someone integrate and acclerometer into the gyro reading to correct the problem
Mark
Brent S. wrote:

Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Have you looked at the Autopilot project? They have kalman filter code for combining the output of an accelerometer and a rate gyro(AVR code). I think this is the basis for the rotomotion products. The code is available at rotomotion as well.
Take a look:
http://autopilot.sourceforge.net /
http://rotomotion.com/index.htm
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
my newsreader really suck, missed a lot of this thread...
Just in case it wasn't covered, a lot of the issue with gyros, is thermal instability. since they output a rotational rate, integrating it is tough enough. Add to that the fact that they drift with temperature, and it is really difficult.
A solution is to put a temperatue sensor in with the gyro, or thermally stabilize the gyro. Better still, is to characterize the gyro through rate and temperatue.
I have heard of better success by low pass filtering, then integrating the signal.
mike

determine
over
the
this
you
USB.
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

A few comments.
1) Gyro's do have drift error that accumulates over time. 2) The earth is moving in space and gyros detect this motion even if your vehicle is sitting still. 3) Use of multiple instruments to track movement can lead to more accurate results. The widely used method for combining multiple measurements in navagation systems is to employ a Kallman (spelling?) filter.
The above summary is almost everything I know, or think I know, about gyros and navigation (not much is it).
TC
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
TC wrote:

I know even less, but a buddy of mine works for Sperry Marine and gave me a tour of his plant and showed some of the navigation gear. They use gyros and three sets of fiber optic motion sensors that look from the outside like a PC hard drive. Inside are many turns of fiber optic cable, and it detects phase shifts as the loop of fiber rotates.
The good news was that the fiber optic sensors are commerical, made by some part of Honeywell. The bad part was that his shop got $50,000 on up for the sensor, and it was six foot high and a foot arround.
--
Pat



Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
"Pat Farrell"

I believe they have similar sensors in a smaller package (but it doesn't mean that it is going to cost less)... they are very expensive
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload
mark,
one approach you might consider is having your robot continuously compare the direction rate measurements from the wheel encoders and the gyro. when they're pretty close, you use the wheel measurements, but when they differ substantially, you use the gyro. i think you'll find that for indoor robots and cheap gyros, wheel encoders will have much less drift over time but this way the gyro will detect when there's an exceptional odometry event such as driving over debris.
that said, no continuously integrating system will stay accurate for very long; you might look into various slam algorithms, though they tend to require a little computer science and a lot of computer.
also, the accelerometer approach works when the gyro is used to measure pitch and roll, not yaw. (my balancing robot uses one for that) unless gravity goes sideways, it will be of no use to you.
chris
Add pictures here
<% if( /^image/.test(type) ){ %>
<% } %>
<%-name%>
Add image file
Upload

Polytechforum.com is a website by engineers for engineers. It is not affiliated with any of manufacturers or vendors discussed here. All logos and trade names are the property of their respective owners.