Comment by Rhett "Wreck" Phillips on February 19, 2010 at 3:52pm

Very cool, I wish MATLAB wasn't so expensive. I'm not familiar with AVL, I'll look into it. Thank you for the post!

Comment by Borja Martos on February 19, 2010 at 7:46pm

To compile EKF.c in MATLAB using MATLAB's

mex -setup

Select a compiler:
[1] Lcc-win32 C 2.4.1 in C:\PROGRA~1\MATLAB\R2008a\sys\lcc\bin

I had to create an EKF.h fileEKF.h

Comment by Borja Martos on February 19, 2010 at 7:54pm

Thanks for the code Paul!

Have you done any work integrating additional states into the EKF to obtain Position and Velocity in NED?

Comment by Paul Bizard on February 20, 2010 at 2:41am

Hi Borja,

Thanks, I forgot to talk about the selection of the compiler.

No I did not work on integrating additional states. But I will one day.

Today I own a UAVDevBoard with "DCM inside". It is a good product.
But I am willing to assemble/buy a more powerful board in order to play with this EKF and with one integrating Position & Velocity. I think this is the only way around because GPS is not accurate and is not "loss of signal"-proof.
The EKF I uploaded is very similar to Bill Premerlany's DCM in the sense that its structure is the same: I used a quaternion to miniminize the dimension of the state vector (4 states quaternion and 3 gyro biases = 7 states) but the correction is based on the same Roll-Pitch / Yaw decoupling and even the expression for the RP and Y deviations are the same. The RP & Yaw error vectors are essential for the accuracy and robustness of the identification process. (Bill P. has a powerful 3D/algebraic vision).

As you can see I used the GPS cog for yaw correction, exactly as DCM does, and I delayed the quaternion before correction in order to phase the two cog (GPS and estimate). This is also done in DCM, anyway in case of the EM406.

I want next to use a magnetomater rather than GPS for yaw estimation, so I will have modify the yaw correction subfunction.

Comment by Paul Bizard on February 20, 2010 at 2:53am

Thanks Rhett,

The interesting point is that AVL is relatively accurate and with this simulator you can tune your gains, at least for the first flight.

The process of computing the airplane coefficients with AVL is not straightforward for a newbie, but is worth the pain.

Developer Comment by Pete Hollands on February 20, 2010 at 3:29am

Here is the link for Mark Drela's AVL program for those that are new to this software.

Developer Comment by Pete Hollands on February 20, 2010 at 3:35am

Paul,

you wrote: But I am willing to assemble/buy a more powerful board in order to play with this EKF and with one integrating Position & Velocity.

Currently the UAV Devboard is using 6% of it's CPU power for the latest version of MatrixPilot at the head of the trunk of development. So there is a lot of spare CPU power to integrate position and velocity. I know that Bill is starting to think now about returning to a "Dead Reckoning" model for position, with the GPS re-calibrating the "Dead Reckoning". This will mean that we know our position every 1/40th of a second.

So I believe we will be undertaking more integration of velocity in MatrixPilot on the UAV Devboard quite soon.

Pete

Comment by Paul Bizard on February 20, 2010 at 4:25am

Hi Pete,

Yes, I've read about this on UDB Google Group, but I want to follow the EKF path and I beleive in the simple airspeed measurement rather than its estimation. The airspeed sensor is not that expensive and I favor simple solutions.
I also think that a magnetometer is required for a rock solid yaw estimation. (What will happen if a gust of wind reverses the speed for 20 seconds ? The estimated yaw will have 180° of error ?). I am concerned about a catastrophic series of events.
Nonetheless, what has been -and is being- achieved by Bill and all of you with such a simple cheap board is incredible. I am also happy to have helped Bill a little. I will fly FPV this summer with the UDB.

Your latest developments are outstanding and together you make up a great team.

Developer Comment by Pete Hollands on February 20, 2010 at 5:13am

"I am also happy to have helped Bill a little." Thank you for your contributions. They were an absolutely vital part of the early success of DCM and the UAV DevBoard.

Developer Comment by Mark Colwell on February 20, 2010 at 6:28am

Bill, Paul, & Pete,
Thank You, for all the fine documentation and clear discussions on DCM, I have learned a boat load, and stand on your coat tails. I am working on ArduPilot IMUv1 , and very near to good high speed flight with neutral stable Stryker II I hope to win SF competition this year.

Comment

• ‹ Previous
• 1
• 2
• 3
• …
• 9
• Next ›
• Page