The issuses of conversion Quaternion to Euler

[chestnut]

Hi all,

For conversion Quad to euler :
*yaw = atan2f((2.0f * (q0 * q1 + q3 * q2)), (q3*q3 - q2*q2 - q1*q1 + q0*q0));
*pitch = asinf(-2.0f * (q0 * q2 - q1 * q3));
*roll = atanf((2.0f * (q1 * q2 + q0 * q3)) / (q3*q3 + q2*q2 - q1*q1 -q0*q0));

have some issuses
asin returns the angle between -π/2 and π/2 radians (equivalent to -90 and 90 degrees)
atan returns the angle between -π/2 and π/2 radians (equivalent to -90 and 90 degrees)
atan2 returns the angle between -π and π radians (equivalent to -180 and 180 degrees)

If pitch or roll > 90 or <-90, it's will wrong value. Can we fix this issuse?

[srinath]

Take a look at
navUkfQuatExtractEuler

The sequence of rotations is important. YPR or RPY or something else.
I think its YPR on AQ.

Also important is whether the quat corresponds to world to Body or Body to world frame.
I don't remember the convention in nav_ukf

Bill can confirm

[bn999]

Can you come up with an example quaternion which fails this conversion to demonstrate the problem?

[chestnut]

[bn999]

Sorry, I meant an example 4 digit quaternion which when passed through this function results in an invalid pitch, roll or yaw. I don't think there is one.

[srinath]

The function should specify rotation sequence(ypr/rpy etc) at least in a comment. Otherwise its confusing IMO

[chestnut]

Hi Bill

For *pitch = asinf(-2.0f * (q0 * q2 - q1 * q3));

Example:

sinf(PI/4) = sinf(PI/4 + PI/2) = x
-> asinf(x) = PI/4 or 3PI/4

[srinath]

[chestnut]

[srinath]

Yep. So we should definitely use only Quats for control.
Anything that passes through the singularity at pi/2 is doomed to fail

Btw, where does this formula come from i believe it should be slightly different.
Look at page 45, eqn 3.63