On Friday, 23 October 2015 00:27:41 UTC+3, wrote:

bruce, i think you might be assuming the butterfly is merely doing dead reckoning, but what it's doing is more complex than that. it uses an extended kalman filter (EKF), which is a type of software algorithm that is good for estimating the state of something that cannot be directly measured (the motion of the glider, in this case). the EKF takes various data steams as input: accelerometer, gyroscope, static pressure, pitot pressure, magnetometer (compass), and time. what it outputs is an estimate of the motion of the glider.

now, if the EKF only had acceleration and time as inputs, it would need an extremely accurate accelerometer and clock, as you have surmised. acceleration is relative, so even small errors in acceleration will snowball over time into large errors when doing dead reckoning this way.

fortunately, the EKF has much more data to work with. since it also has access to measurements that are absolute (heading, altitude, airspeed), it can integrate all the data in a way that makes the accelerometer error much less significant. since acceleration is only one piece of the puzzle, it has less influence on the overall motion estimate, and therefore the drift you would expect with dead reckoning is diminished.

thanks to the power of the EKF, an extremely accurate IMU (accelerometer, gyro) is not necessary. i wouldn't be surprised if the butterfly has an IMU of similar quality as that of a smartphone.

Kalman (and EKF) is a process that you run into pretty much everywhere. It's quite elementary way of correct predicted state of something on basis of comparing previous predicted state to current observations. It doesn't do theh inertial navigation itself, it's just way of polishing things numerically. For example, if airspeed measurement is systematically too small, Kalman filter corrects that error (this is maybe too elementary example, reality is more complex for sure). I think important part of achieving this state of inertial navigation (without GPS) is by using measured TAS.