Re: Accuracy of Omni drive?
 

Here's a clip of 2077's first effort with a three wheel omni, back in 2010.

http://www.youtube.com/watch?v=zPlMZlfHg2o

There's no footage of it driving straight, but it did that just fine, with a very simple proportional feedback from the gyro in the main drive loop, applied only when the rotation control input was zero. The gyro was also used for field-relative driving, what was really being demonstrated in the video.

A couple of amusing robot behaviors we saw in this machine:

• If you walked up to the robot while it was enabled and sitting still, and rotated it by hand, it would return to its original heading by itself.
• Occasionally we would put the robot down for a test and it would slowly rotate for no apparent reason. We eventually figured out that we'd been powering up the robot while people were positioning it for the test. If the robot was moving during gyro calibration it had a different notion of "no rotation".

Of more practical significance we learned:

• Simple gyro feedback is fine for solving the "drive straight" problem, either with omnis or, as we later learned, mecanums. We later learned to use encoders, but they're not essential.
• The small omni wheel roller radius adds a lot of friction to sideways motion on carpet, and with this configuration at least two wheels always have a sideways component at any time. This robot was unacceptably sluggish in competition.
• Three wheels are a perfect solution to uniform wheel loading, but with only three CIMs that could never all be fully applied to linear motion at once, it was way underpowered.

On the balance, I like mecanums better, because the friction losses seem smaller (larger roller diameter) and more importantly, not so uniform in different directions. The less the rollers are turning, the less power they waste, so forward/back motion is much more efficient with mecanums. You lose it back when moving sideways of course, but you normally do that less.