Rotate yaw in position control mode,the horizontal velocity is wrong with gps measured velocity.

[garlinplus]

No description provided.

[garlinplus]

the log address is https://logs.px4.io/plot_app?log=912ad7f9-74e4-45c1-9a74-5a6072930797

[CarlOlsson]

That's a very interesting logfile, the IMU biases diverge during the rotation. Do you think you could collect a replay logfile during the same maneuver?

[garlinplus]

the IMU biases diverge during the rotation

The gyro and acc bias seems to be large than formal flying log as followed:

[garlinplus]

That's a very interesting logfile, the IMU biases diverge during the rotation. Do you think you could collect a replay logfile during the same maneuver?

The imu bias is larger than formal flying,but Even if as that,the gps and mag measure can not correct those with ekf measure update process?

[abishekvupputur]

I notice that your yaw angular rate is about 100 deg/s. I think it is quite possible that you are measuring centrifugal acceleration in the body reference frame. Due to lack of observability of centrifugal force, in body frame it is interpreted like a "added bias (function of spin) during large yaw rates". So it makes sense that the bias estimates in [x,y] is diverging to accommodate for this error caused as a consequence of spin. If this is the case, I think the EKF is working perfectly to accommodate this. There 2 possible ways to get rid of this problem.

1. Centre your IMU closer to the center of gravity of the drone. The closer you are, lower is your "acceleration bias divergence".
2. Write a small calibration routine, manually calculate the offset of imu from the centre of gravity and calculate your acceleration caused as a result of spin before passing the values through your EKF. This compensation for the added acceleration can be done by subtracting a_c from the measured acceleration from IMU:

a_c = ω x ω x r + ω_dot x r.
a_c ~ ω x ω x r
where,
a_c - calculated acceleration in [x,y,z]
ω - body angular velocity in [x,y,z] (gyro bias corrected values)
r - imu offset from cog in [x,y,z]

Note: I ignored the noisy angular acceleration part here ω_dot, the gyroscope values are noisy enough and the derivative can make it worse.

Does this fix your error?

[abishekvupputur]

I think this is also a potential duplicate of: #621

[priseborough]

This looks like the issue that was fixed by #1026

Please re-open if not fixed for you.