I have raw accelerometer data (w_ib^b) and magnetometer data from a UAV test flight. I would like to reconstruct the quaternion dynamics of the UAV without accumulating error from integration.

The method I was using involves solving the following integration: d(q_b^w)/dt = 0.5*Omega(w_ib^b)*q, where Omega(w_ib^b) is of the form: [-skew(w_ib^b) w_ib^b; -w_ib^b' 0].

I was wondering whether there was an alternative method of computing the aircraft attitude given raw angular velocity and magnetometer information from a UAV?

  • 2
    $\begingroup$ I think this is of topic here. Other than coming from a UAV, this is more suited to mathematics.se or physics.se. $\endgroup$
    – Ron Beyer
    Aug 10, 2017 at 21:58
  • 4
    $\begingroup$ It's flight dynamics. Quaternions are used to work out the position and attitude of aircraft in earth reference axes. Euler angles are more intuitive, but if the aircraft is pointing straight up the Euler angles are undefined and quaternions are the only solution. $\endgroup$
    – Koyovis
    Aug 11, 2017 at 1:11

1 Answer 1


If all equations are digitally computed multiple times per second, all variables are updated at every frame and the integration error is only one time-step old. Full Flight Simulators use four sets of quaternions to compute the state of the aircraft in earth reference frame. To illustrate how it is done in FFS flight dynamics, here is a partial set of the steps that take place:

$$ \dot{E_1} = 0.5 \cdot (-E_4 \cdot p - E_3 \cdot q - E_2 \cdot r) + E_1 \cdot C_Q * C_{Q\Delta t}\tag{1}$$ $$ E_1 = \dot{E_1} * \Delta t \tag{2}$$ $$ \Sigma E^2 = {E_1}^2 + {E_2}^2 + {E_3}^2 + {E_4}^2 \tag{3}$$ $$ C_Q = 1 - \Sigma E^2 \tag{4}$$ $$ E_{1P} = E_1 * \frac {1}{\sqrt{\Sigma E^2} } \tag{5}$$

  • Equation (1) is one of four equations for $\dot{E_1}, \dot{E_2}, \dot{E_3}, \dot{E_4}$, while $p,q,r$ are aircraft roll, pitch, yaw rates.
  • $C_Q$ is a quaternion correction factor which uses the sum of all quaternions squared.
  • $C_{Q\Delta t}$ is a timing correction factor and should be less than the iteration rate in Hz. A constant value is empirically chosen that gives good stability.
  • Equation (2) is a simple Euler integrator using the actual time step of the computation frame. So if we're computing at 1000 Hz, $\Delta t$ = 0.001.
  • $ E_{1P}$ is the prime quaternion substituted in the transformation matrix from body to earth axes.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .