IMU without gyros

[aitor_ol]

Hi all,

I have been reading this topic for a while and i am very impressed about your work !!

I want to develop my own autopilot system and i am considering to use just several 3D acelerometers (7€) instead of the more costly gyros (30€ ??) to calculate the attitude of the plane.

The basic idea is to calculate the rate of orientation change by means of aceleration diferences of the acelerometers. Sensitivity would be better if the acelerometers are as far as possible each other. For example, two vertical sensing accelerometers located at the wing tips could provide roll information.

What do you think of it ??

regards.

[Joan Carles]

Hi,

I'm also developing my UAV and about the same doubt with gyros and accelerometers.
Certainly, if put at the wing tips, angular acceleration is converted to linear acceleration and amplified, I think it's a good idea, at least in the sketch.

I thought about using just one 3-D accelerometer to measure attitude by measuring the gravity vector with respect to airplane body. But the problem is accelerometers also measure real acceleration by the propeller or other reasons and the system would 'think' it's pitched down...

Your idea seems ok. Vertical differential accelerations means a roll, horizontal differential accelerations a yaw, and so on. Average accelerations of the accelerometer pair would mean linear acceleration.

I think it's a good, money saving idea --- :)

[jetblackaircra]

The concept is good, but in practice I think you'll find that it will be difficult to impliment. First of all the accelerometers are noisy and when you have two sensors the amount of noise just doubled. Also, you would need two sensors which were very well matched and calibrated to each other. A third problem would be that you are now having to integrate twice to get attitude. You're measuring angular acceleration, not angular rate. Every time you integrate you lose accuracy. A sensor that directly measured attitude angles would be best. Second would be a sensor that measures attitude angular rate. Third best would be a sensor that measures angular acceleration, which is what you would be doing.

Good luck.

[aitor_ol]

I agree about the usefulness of integrating twice the measured accelerations to compute the attitude of the plane, drift, not well paired acelerometers, and so on.

About the noise of the accelerometers, perhaps its not an issue. When we integrate a signal, the high frecuency noise gets smaller. I guess that the double integration should provide a noise free signal, altough the drift could be very high.

but there is one thing we can try,

we can take advantage of the dc response of the accelerometers to obtain a very low frecuency signal representing the acceleration of the gravity, i have heard about this somewhere but i dont remenber where.

This signal could be feeded to a kalman filter based fusion scheme to eliminate the drift of the attitude computation.

[jetblackaircra]

You are correct that integrating becomes sort of like a low pass filter in that the noise will no longer appear in your signal, but the noise gets added into the integration to cause drift. There is already a problem with excessive drift when using angular rate gyros which are very accurate and noise free. The added integration will increase that drift even more.

I'm not saying it won't work, I'm only saying it will be difficult to get it to work properly and accurately.

[rblilja]

It indeed works, maybe not as good as gyro-based solutions, I can't determine that.

Interesting reading:
http://www.soe.ucsc.edu/~elkaim/Documents/QuatAttitude.pdf
http://mtc-m16.sid.inpe.br/col/sid.inpe.br/mtc-m16@80/2006/08.09.17.35/doc/DINCON-2006_Edmundo_Marques-Final.pdf

There are tons of papers of this subject, gyro-free IMU.

[aitor_ol]

Thank you for the links rblilja.

regards