Bitcraze Forums

Hey guys!

I am currently designing a nonlinear trajectory controller for my crazyflie 2.0. It requires position data from the quadcopter so my question is concerning the kalman filter implementation.

Is it possible to acquire accurate position data from the kalman filter without any additional external unit, the loco positioning for example, and if so, how accurate are the measurements? I know that positioning without any extrinsic system is a really complex problem and the integration of acceleration data leads to massive errors in the position (using the euler method f.e.), however i am still curious about the precision of the kalman subsystem.

Thanks for any helpful replies in advance!

Hi, you can compile the Crazyflie with Kalman filter (by writing "ESTIMATOR=kalman" in tools/make/config.mk) and observe the position estimate to see for yourself. My previous observation of what is happening is that at first it is fairly stable, then it starts drifting and the position ends up hitting the hard-limit of the EKF position estimate: +/-10m. Unfortunatly without any position or velocity information the double integration of the accelerometer is just diverging very quickly (in seconds).

Hi!

I've tested the kalman function a few months before and the results seemed completely off when i moved the copter in my hands, hence the question. The good news is that i only need the values just for a few seconds when the copter moves from a position into a different one. Do you guys have information about any projects whch tried to take this direction (positioning without external units)?

Hi,
I think you will still find it completely off, it has not become better in the last few month with using only the accelerometer.

We are about to release an optical flow deck (wiki) that allows to track velocity quite well, in good condition there is virtually no drift in position. This is one way to track position without any external systems.