Re: [Paparazzi-devel] Booz attitude estimation

[Felix Ruess]

Hi Marko,

the assumption is that the quad is not accelerating (which is also
true if you fly at constant speed in one direction). In this case you
can estimate the roll and pitch angles from the accelerometer
measurements, meaning the body frame sensor data ax and ay is relevant
for the attitude estimation. Of course it is correct that during
manoeuvres when the quad is accelerating the estimate from the
accelerometers will not be correct. But on the long run they provide
useful information as cannot accelerate indefinitely in one direction
and usually don't fly in tight circles all the time. This is why the
simple filter "fuses" (high passed) gyro data for short term and (low
passed) accelerometer data for long term estimation.

Hope this helps to clear it up a little.

Cheers, Felix

On Sun, Jun 6, 2010 at 7:15 PM, Marko Thaler <address@hidden> wrote:
> Hello Poine,
>
> Thank you for your answer.
>
> I understand and agree with everything you wrote but would like to point out
> the following concern.
>
> Xdotdot = 0 assumption leads us to believe we can estimate the roll and
> pitch angle based on the body frame [ax ay az] IMU measurements. In
> quadrotors Xdotdot = 0 can only be achieved in perfect hover (roll and pitch
> = 0) or if we hold the quad in our hand and rotate it to a certain attitude.
> If the quad is flying, non-zero attitude angles (roll and/or pitch) will
> cause the quad to accelerate in a given direction. The gravitational (F_g =
> m g) and inertial force (F_i = m a) will cause our accelerometer
> measurements in body frame to be [0 0 az]. This measurements assume IMU in
> the center of gravity, no external forces acting on the quad (wind, etc.),
> zero sensor noise and motor vibration. The only acceleration in the body
> frame we can measure, based on the previous assumptions, is caused by the
> motor/propellor thrust force. This force is always parallel to the az
> direction no matter what the attitude angles of the quad are!
>
> This leads us back to my basic concern that we are trying to estimate the
> attitude angles based on attitude non-relevant body frame sensor data ax and
> ay (non-zero values are caused by sensor noise, motor vibration, external
> forces, etc.).
>
> Please, let me know if I can improve the clarity of my concerns with some
> drawings and formulas.
>
> I am looking forward to hear your thoughts and opinions.
>
> Thank you and kind regards,
>
> Marko
>
>
> On Jun 5, 2010, at 12:01 AM, antoine drouin wrote:
>
>> Hi Marko
>>
>> What the accelerometer senses is the non gravitational acceleration,
>> in the inertial frame, and expressed in the sensor frame... that is :
>> [ax ay az] = Xdotdot - g
>> What the filter does, is assume that Xdotdot is zero ( which is more
>> or less true for a hovering vehicle ). Then the accelerometer is only
>> sensing.... the opposite of gravity. Knowing that the gravity is [0 0
>> 9.81] in inertial frame, you can express the first two euler angles as
>>
>> phi = -atan(ay/az)
>> theta = asin(ax/|a|)
>>
>> (  use of atan and asin is in order to constrain phi to -180:180 and
>> theta to -90:90, but it's not important )
>>
>> the formula you show in your post is just a linearization of that.
>> Also, if you're trying to understand the value of ACC_AMP, please note
>> that i did not compute the slope of the linearization at the origin,
>> but at pi/8 ( so that the value is exact at pi/8)
>>
>> It's a bit hard to explain all that without drawings and formulas. If
>> that doesn't do the trick, ask again and I'll make some drawings.
>>
>> hth
>>
>> Poine
>>
>>
>>
>>
>>
>>
>>
>> On Fri, Jun 4, 2010 at 11:37 PM, Marko Thaler <address@hidden>
>> wrote:
>>>
>>> Hello everybody!
>>>
>>> I am building my first Booz quad and studying it's code. One thing that
>>> puzzles me is the implementation of the Complementary Euler Attitude
>>> Filter.
>>>
>>> On lines 159 and 160 in
>>> sw/airborne/booz/ahrs/booz2_filter_attitude_cmpl_euler.c we estimate the
>>> body attitude angles based on the accelerometer measurements:
>>>
>>> measurement.phi   = -booz_imu.accel.y * ACC_AMP;
>>> measurement.theta =  booz_imu.accel.x * ACC_AMP;
>>>
>>> Quadrotor flight dynamics dictate that x and y acceleration in the body
>>> coordinate system (imu) are always zero. Non-zero values are only
>>> produced
>>> by sensor noise, external wind force, IMU not in the center of gravity
>>> etc.
>>> Additionally, IMU measurements can directly be used for attitude
>>> estimation
>>> only when the system is not accelerating (in hover or constant speed
>>> condition). If we ignore the extreme cases this condition can only be
>>> achieved when phi and theta are 0.
>>>
>>> Based on this I can not see the value of direct attitude estimation based
>>> only on the accelerometer data.
>>>
>>> I would really appreciate if somebody can explain the purpose of this
>>> part
>>> of the code or show me if I have missed something.
>>>
>>> Thank you very much for your help!
>>>
>>> Kind regards,
>>>
>>> Marko Thaler
>>>
>>>
>>> _______________________________________________
>>> Paparazzi-devel mailing list
>>> address@hidden
>>> http://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>
>>>
>>
>>
>
>
> _______________________________________________
> Paparazzi-devel mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>