Optitrack and Crazyflie sloppy takeoff

[alwin]

We are working on a setup in which we try to hover the Crazyflie 2.0 by using Optitrack for tracking. Optitrack Motive is sending out coordinates via VRPN and we are running an Ubuntu installation with a VRPN Client which is translating the Optitrack coordinate system to standardized coordinates. RoboViz is showing the motion paths very accurately and in line with tracking in Motive. However, CF's response is not quite accurate.

As you can see in this video the take off is very sloppy. CF seems to ignore the tracking system and takes off in a seemingly random direction. After a while it seems like it's suddenly using the tracking data and directs itself back to the target location.
https://www.youtube.com/watch?v=sGau9ur ... e=youtu.be

Tracking is equally acurate in the whole setup and RoboViz is showing correct paths so it doesn't seem to be a tracking issue, more of a coordination issue on the CF.

Any pointers or insights would be greatly appreciated

[whoenig]

Which software framework are you using? If you want good tracking performance, I recommend the Crazyswarm project (https://github.com/USC-ACTLab/crazyswarm). It requires a custom firmware, but runs everything onboard (including a non-linear controller) and can deal well with packet drops.

[larry12193]

What are you using to takeoff with? I had similar issues when just commanding a gradually increasing thrust. These issues disapeared when I started sending a hover setpoint (vx, vy, yawrate, zDistance). This was also while making the on board estimator default to the kalman filter based estimator. Commanding a zero vx, vy, and yawrate will make for a more stable takeoff while gradually bringing up the zDistance setpoint. That ground effect really makes the quad float like a drop of water on a hot frying pan.