Hi everyone,
Having been quite busy with work in the last few months I finally found time to write a new blog article about the latest developments in the UAVP-NG project.
In my last blog entry I announced that we will open the UAVP-NG shop soon, and so we did.
On the 26. December of 2009, the first day of the 26c3 - The 26. Chaos Computer Club Congress - we opened up the project's non-profit UAVP-NG shop to the public. We did not anticipate such a high demand and so we were sold out on the 13. January this year for two weeks... ;) We learned our lesson and increased our stock accordingly.
Having rolled out our hardware to the public we finally found time for some source code cleanup and further development.
End of January 2010 we released NGOS-0.55 with support for the VDRIVE2 and the new X8 and X8X coax HAL and the ng-tools source. End of March NGOS-0.56 which included a new test HAL called 'blc' to allow users to test single BLCs. In April NGOS-0.57 and NGOS-0.58 followed with rc-mode support, yaw stability improvements and a yaw filter. And finally, in June we released NGOS-0.59 which introduced the new native S3D 2.4Ghz support for the S3D receivers from ACT which Ralf and I implemented in spring.
In the meantime I pondered some new papers I've read on the DCM - The Direction Cosine Matrix and it's applications for UAVs. Several very smart folks wrote some heavy papers about the stuff and having read a lot about it in the last months I decided to try to implement a new DCM based controller, which I called amir-ng.
The basic idea of a DCM controller is to remember the attitude and orientation not by using 3 euler angles, but by using a DCM - a direction cosine matrix.
It's one of the 3 often used methods to remember attitude and orientation. Besides Euler angles and a DCM there are Quaternions to do the same. Mathematically a DCM and Quaternions can be shown to be identical except for the number of storage needed and some differences in mathematical complexity. Quaternions use 4 variables to describe the state, a DCM uses 9 variables, of which 4 are free.
The DCM essentially consists of the 3 axis vectors of the two involved coordinate systems, seen from each other. It will store the attitude and orientation, but needs to be corrected for mathematical deformation and for drift effects happening by the gyro integration and errors introduced.
This drift correction of a DCM can be done in different ways, similar to the Euler case, where you can use different Kalman filters or complementary filters to do the job of fusing acc and gyro information.
Contrary to Euler angles the DCM's state space resides in the SO(3) Manifold - the special orthogonal group of rotation - and there, the rotation problem is linear! This has several important implications...
As you probably know a linear Kalman filter is a optimal solution if, and only if, the problem is linear. Otherwise you have to resort to an extended Kalman, which tries to linearise the curve at the point of interest by using a first order Taylor approximation or similar solutions which essentially work, but also inherently fail to be optimal, by their way of working (eg. a first order Taylor approximation simply no longer is optimal).
So essentially you can't get a optimal solution using Euler angles. And the Euler angles singularities always will irritate you... g
Now guess what happens if we look at the problem in SO(3). As a rotation is a linear operation here, we can apply a simple linear Kalman filter to the problem and get an optimal solution. This is very helpful as a linear Kalman filter is implemented much more easily than an extended one.
So using three single linear Kalman filters to guess the drift compensation in SO(3) results in an optimal solution and I tried to implement this in the amir-ng controller.
The result seems to come out quite satisfying! Take a look for yourself on the following short movie showing the maidenflight of the DCM controller with Kalman drift compensation:
Download: mp4 (480x360)
Play on: YouTube
Parallel to my work on the new controller, Stefan finished cleaning up the compass support on the SBCTRL side. While he was working on implementing compass and GPS support in the older amir controller, I decided to implement GPS and compass support in the amir-ng controller. This made sense as the application of the compass heading for drift compensation is different in the two controllers. Instad of using a heading over ground derrived from GPS measurements as used in many DCM implementations, we use the attitude compensated compass measurement for yaw drift compensation in the amir-ng controller.
Having implemented yaw drift compensation using a compass, GPS navigation suddenly was very easy to archive. As a DCM contains the coordinate axes of the two involed coordinate systems, it's very easy to compute heading deviations and courses. It's simply vector geometry...
Realizing that, a simple GPS Position Hold algorithm was hammered together and tested out, some weeks ago! This was the first successful test on the 24. June of 2010:
Download: mp4 (480x360)
Play on: YouTube
Again we use 3 linear Kalman filters to fuse acc accelerations and GPS position.
Having realized a first PH flight, we started preparations for the 3. UAVP-NG Multicopter and Developer Meeting here in Zurich, Switzerland.
It took place on the 9. - 11. July of 2010. Here's some footage on the event... I think we never had so many different NG flight controls in one place... :)
You can find some impressions of the event here: 3. NG Developer Meeting in Zurich (2010)
UAVP-NG - 3. Developer Meeting in Zuerich on the 9/10/11. July 2010 on Picasa
Rob's movie:
3. NG Developer Treffen from Rob Robot on Vimeo.
And Ygramul's movie:
NG Treff 2010 in Zuerich from Ygramul on Vimeo.
I would like to thank all visitors for coming by! It was a lot of fun, as always!
So, I hope I got all the News out... ;)
Last but not least I was 2 weeks on diving holidays, but I will write another blog entry about that... :D
Cheers!
- Amir