Fig. 7. Predicted versus measured change in roll during the four phases of the wingbeat cycle. (A) Mid-upstroke to the start of downstroke, (B) the start of downstroke to mid-downstroke, (C) mid-downstroke to the end of downstroke, (D) the end of downstroke to mid-upstroke. Because the inertial predictions do not take into account any initial roll velocity, we high-pass filtered the measured roll angles with a cut-off frequency of 3.5 Hz prior to computing the change in roll for comparison with the inertial predictions.