Working with an IMU....

[panzapart]

Hello everyone,


I am working on a project in a team and we are constructing a vehicle for all terrains. We want to have odometry measurements from encoders combined with an IMU. So we are facing the double integration problem and the IMU errors. I guess you understand what I mean...

So I want to ask you if you have experience and if you can tell me which is the best accurate way to do the integration. Of course I think I have to use Kalman filtering for the data.

Thanks in advance!

[Admin]

It's actually a somewhat complicated problem.

What you need is something called a 'Kalman Filter', used to take info from many sensors and combine them into something that makes sense.

http://en.wikipedia.org/wiki/Kalman_filter
http://www.cs.unc.edu/~welch/kalman/
http://en.wikipedia.org/wiki/Extended_Kalman_filter

You can find more by searching this forum.

[panzapart]

thank you for replying  : )

the use of kalman filtering is one of our intentions but fisrt,were are facing a strange  problem in double integration..

as we take the accelerations (m/s^2) and we integrate in very short time periods ( equals to the clock of the imu (180Hz=5,5ms) ) during a short period of time ( the experiment is just x-axis movement with constant speed of about 0.5 m/s ) , our results vary from values 0.05 to 0.06.this means an average 10 times smaller value for velocity..the type of waveform is smooth as a constant velocity,but the value is small..this is our main obstacle..the integration should give values much bigger as there is accumulated error during the process.
the kalman filtering we are planning to use will make a better estimation as me assume,but i think the problem is somewhere else..

is there anything vital that we should take into consideration about the acceleration values..

some details..the accelerations are given in the navigation frame,the method of integration is calculating the time slot and multiplying with the accel value ( we also can add the small triangle lost in the procedure as (acc(t+1)-acc(t))*dt/2 ,but there seems to be small difference..)..any ideas would be much appreciated..thank you in advance

[hgordon]

Accelerometers completely useless for fine measurement on mobile platforms with any vibration.  If integrated over a longer period of time, they will be useful as tilt angle sensors, but you won't extract any useful odometry information.  You will have somewhat better luck with a yaw gyro in detecting direction changes, especially if you can integrate that data with compass measurement.

If  you are operating outdoors (implied by "all terrains"), you should integrate an absolute position sensor such as GPS, and use that to help correct odometry errors over longer distances.

[Admin]

Hmmm I didn't understand your question . . . also, I'm no Kalman filter expert for sure.

I'll be working on a project that requires Kalman filters in about 6 months, so maybe then . . .

Accelerometers completely useless for fine measurement on mobile platforms with any vibration.  If integrated over a longer period of time, they will be useful as tilt angle sensors, but you won't extract any useful odometry information.

Well, if you characterize the vibration and its semi-predictable, or vibration is much less than vehicle acceleration, then it'll still be useful. But you need to know the vibration frequency and amplitude to filter it out . . .

Otherwise, I agree with hgordon on all points.

[panzapart]

thank you for the quick reply..i know that our problem is very specific and hard for an external observator to understand..all our indications show that what with an imu there is no reliable way of measuring velocity or position..

to be more clear,what we are trying to do,is to get in very small periods of time,instant velocities with one integration ( to be given for slam localisation purposes)
and the problem is that the value of the speed produced is 10 times smaller..(it is estimated  0.45 m/s and we calculate only 0.042..) this is the odd thing that is driving us crazy..any ideas..any...
thank you for reading : )

[ArcMan]

Are you sure you don't have a sensor scaling problem?
Have you taken readings from your accelerometer when oriented with gravity to see that your getting readings of ~ -9.8 m/s^2?