Gaining VIO altitude estimates only in one direction

[hmlow]

Hi devs,

Im stumped and really need help.
I've shifted all the hardware from a Seeker into my own airframe, and have done the following:
-configured the extrinsics
-re-calibrated IMU
-re-calibrated tracking camera
-the new airframe flies well in Manual/Stabilized mode

I have yet to switch to Position mode because - while on the ground (motors not spinning; no vibration etc), when i move the aircraft in the positive x-direction (reference frame where the VIO initializes - pointing fwd towards the nose of the aircraft), the z-position estimate decreases (ie it thinks it is gaining altitude).

The same phenomenon happens regardless of the aircraft's orientation/heading as long as i move towards the x-direction. The Z position estimate moves by -0.3m for approximately 5m moved in X direction. X and Y positions are estimated correctly

As i move back towards the starting pt (negative x-direction), the change in z-position estimate reverses and it goes back to 0 (almost) at the starting point. Again, this is regardless of the aircraft orientation.

Im stuck... help needed and appreciated.

Thanks

[hmlow]

This post is deleted!

[Chad Sweet]

Just FYI, on our reference platforms as altitude increases, z decreases as well. We just rotate it in voxl-vision-px4. Check out this blog post here: https://www.modalai.com/blogs/blog/how-to-autonomously-navigate-with-px4

[hmlow]

@Chad-Sweet actually what I meant was when I moved the aircraft purely in the X direction, the Z estimates decreases.

Sorry if I wasn't clear

[hmlow]

Hi... any of the devs have an idea of what i might have misconfigured?

I've just set up another brand new set of VOXL and tracking camera and im facing the same problem.

Since the issue is systemic, it could be some misconfiguration but i cant figure out what i might have done wrongly

[James Strawson]

How is the VOXL oriented on the new airframe, is it the same as on the seeker?

voxl-qvio-server is only aware of the relation between the imu and tracking camera, and as such outputs its odometry relative to how the IMU is orientated when it starts up, even if that's crooked or upside-down. It also outputs an estimate of the gravity vector in that frame which voxl-vision-px4 then uses to try to correct for any tilt the drone may have when it started up using the imu-to-body relation in the extrinsics config file. If the voxl board is oriented differently on the drone body compared with seeker then that entry in the extrinsics config file will need to be updated too.

To see the raw output of voxl-qvio-server before the gravity and orientation corrections are done, you can run voxl-inspect-qvio. This should output position in XYZ relative to how the IMU was oriented then the drone starts up which in the case of Seeker is crooked due to the front legs.

voxl-inspect-pose -l will output the corrected orientation that's sent to PX4.

[hmlow]

@James-Strawson ah I get it now. There was nothing wrong at all, it was all about reference frames. I should be looking at voxl-inspect-pose -l instead.

Thanks!