IMUs are very fast, though, so if you can work out the right timing, it should be more accurate than the dual antenna, and perhaps react quicker, which would work better on little bumps. I’m anxious to see the code that is being worked on. It’s very exciting stuff. It could be that the secret to getting good and accurate roll data is to figure out the latency between the time the GPS observation was made and the time it comes out of the receiver in NMEA. Then if you cache the last couple hundred ms of IMU readings you can reach back in time and grab the corresponding reading.
For me the main benefit of the double antenna is it’s a bit simpler to set up.