VOXL CAM Fully Integrated Robot Perception System Drift



  • We noticed that our px4 flight core failed the accel cal out of the box on qgc.

    We had to assign the folllowing orientation to the imu in order to pass accell cal and have the proper roll, pitch, yaw on the HUD after cal.

    roll 90, pitch 180, yaw 90

    This system seams to have not been flight tested or atleast calibrated out of the box.
    We believe this system should have been calibrated from the factory.

    We also noticed the setting on the voxel for the stereo camera orientation was inverted as well.

    We ran through the camera configuration promt and inverted the stereo camera to be correct but are unsure what other configuration steps we need to do.

    We saw many configuration scripts but have not found anything specifying imu orientation on the voxel.

    We are noticing large drift (~2cm/s) and would like to know how to remediate this.

    Are there any pre configured files or ways you guys can check our settings?


  • Dev Team

    @Enstrom1 thanks for the suggestions and appreciate the feedback. Will talk with the production team to see if we can implement some changes that you recommended.

    To further help, I will need some information:

    How are you going about adjusting the orientation of the imu? Is that through QGC?

    After inverting the stereo cameras, it should be good to go. Did you invert it using the voxl-configure-cameras?

    What configuration scripts were you looking at? Was it configure-camera or configure-mpa?

    Are you seeing the drift in QGC Mavlink Local Position Ned or in Voxl-Qvio? Have you attempted restarting the service? Does it still happen after a powercycle?
    There is some drift, however its usually a small amount of drift. Vio, if it sits too long or if the tracking camera can't see features, will start to drift or blow up. Your best bet to reset it is to either reset the service or power-cycle.

    As far a pre-configured scripts depends on what you're trying to do. We have parameters in QGC, there's some files in mpa if that's what you're referring to?



  • @Adrian-Hidalgo
    Hey,
    Sorry for the delay on this we were sent on a field deployment. I would like to open this conversation up again. I will have to get back to you with the answers to your questions above.
    We have an interesting outdoor use case for your odometry. Have you all seen degraded performance in dusty environments with large dust clouds being kicked up by the drone on takeoff?



  • @Adrian-Hidalgo
    anything on this?


  • Dev Team

    @Enstrom1, I would have to direct you to either @RichieRich or @Jeremy-Schmidt.
    I do know that VIO needs to be able to see contrasts and objects fairly clearly to work at it's best.


Log in to reply