Re: [Paparazzi-devel] Paparazzi EKF- and compelmentary- filter

[Prof. Dr.-Ing. Heinrich Warmers]

Hi Flex,
our EKF with float has a run time of  7ms without the use of tables for arcosin and  arctan2.
For a normal plane and a quadrocopter there is no problem to run the EKF all 25ms. There are  no timing problems with
the feedback loops since backstepping is used.
See our videos on  http://vimeo.com/10134857
If you can give me the code of EFs with more states   it would help us.

Cheers, Heinrich


Felix Ruess schrieb:

Hi Andrew,

some (unpublished) filters are used on an external more powerful
hardware (gumstix and atom). There is also the error state space KF I
wrote in float for testing in simulation under
sw/airborne/booz/ahrs/booz_ahrs_float_lkf. But you can't fly with that
one at the moment (too long to compute in float and probably still has
some bugs I guess).

Cheers, Felix

2010/4/13 Andrew S <address@hidden>:
  

Hi Felix,
Where is the paparazzi KF?
Andrew


13.04.10, 08:35, "Felix Ruess" <address@hidden>:

    

Hi Andrew,

 Whether you represent your angles in a way that they have
 singularities (like euler angles) or not (like quaternions or dcm) has
 nothing to do with the choice of your estimator (e.g. EKF or
 complementary filter or your PI controller). In any case you have to
 transform your IMU measurements from the IMU frame into the navigation
 frame (assumed to be the same as the inertial frame considering the
 quality of the sensors).
 The paparazzi KF uses quaternions as angle representations (you really
 don't want to use dcm here) but also converts the latest quaternion
 attitude estimate to euler angle and dcm representations for ease of
 use.

 Cheers, Felix

 2010/4/6 Andrew S :
 > Hi
 > For my view, EKF is unnecessary as it loads microcontroller. You can use even simple PI controller. It works worse than EKF, but quite well. For good results PI regulator should be adjusted for 10 seconds of correction after 180 degrees error. Also I can say cheap two axis gyros by ST are unusable because of cross axis interference and whey do not have tolerancy against vibration. Complementary one axis Z-gyros are better.
 > For correct angles EKF is not a solution, you need for example Directional cosine matrices for calculations between body and earth frames. The problem is that gyros works in the body frame, but accelerometers works in the earth frame. In fact when pitch is near 90 degrees, the roll gyro indicates jaw :)  EKF will not helps for this.
 > Andrew
 >
 > 02.04.10, 17:41, "Prof. Dr.-Ing. Heinrich Warmers" :
 >
 >> Hello,
 >>  we made same experiments with a EKF based  on the rotomotion solution in
 >>  combination with the tiny13 and a low cost analogue IMU from sparkefun.
 >>  By the test i saw that the software is not  possible to  deal with
 >>  pitch  angeles over  90°. So  to fly a looping is impossible.
 >>  If  90° arrives it is  impossible to get the correct angeles by the g
 >>  force since static  z acceleration is zero.
 >>  If the pitch angels is more than  120°  the roll angeles rotate fast.
 >>  What is about the  paparazzi  booz ekf and complementary filter. Can
 >>  they deal  all angeles ?
 >>  In the video from 2006  i saw only small  angeles.
 >>
 >>  Cheers
 >>
 >>  Heinrich
 >>
 >>
 >>
 >>
 >>
 >>  _______________________________________________
 >>  Paparazzi-devel mailing list
 >>  address@hidden
 >>  http://lists.nongnu.org/mailman/listinfo/paparazzi-devel
 >>
 >>
 >
 > --
 > Здесь спама нет http://mail.yandex.ru/nospam/sign
 >
 >
 > _______________________________________________
 > 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


      

--
Здесь спама нет http://mail.yandex.ru/nospam/sign


_______________________________________________
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