[EricH]

Using the 100 RPM bias figure that Coffeeism mentioned (note--I'm not sure how accurate that is; someone may want to figure it out), and the free speed of a CIM at 5500 RPM, that's a 1.8% difference from switching directions. Pretty small. When you gear down a CIM, you multiply that a bit--but it's still pretty small. For a free-speed motor, it's entirely possible that one motor and its otherwise identical twin have a larger free-speed difference than a motor and its bias.

Would it work? Probably. Are there other, simpler, and possibly more elegant, ways of doing the same thing? Yep; I can think of 2-3 off the top of my head, ranging from ignoring automode entirely to having a gyro and a pair of encoders and some programming to ensure that the robot maintains heading to within 1/1000 of a degree or some other ridiculous degree.

I think it's a new solution to a much-solved problem; while I can appreciate the engineering, I don't see why it's better than the other solutions.