@AndriiHlyvko The issue is with dependencies. The emulator is built using the actual filesystem used on the VOXL platform so you get all of the correct dependencies. With voxl-cross it is sometimes very difficult to setup the dependencies properly.

@ejohnson Yes, that is a bug. Don't have a suggestion off the top of my head.

@Judoor-0 , I am sorry, but i cannot help you with the mavsdk issue due to lack of knowledge of the subject. Please follow up in the appropriate thread and someone else should help you out!

Alex

@dario-pisanti Hi Dario,

Thanks for sharing your issue. I'm also working with the Qualcomm Neural Processing SDK on VOXL2 Mini.

The error you mentioned could be due to:

Model compatibility issues during conversion.
SDK configuration, especially for GPU.
Double-check the GPU-specific settings in $SNPE_ROOT/bin/envsetup.sh and try running a simpler model to verify the GPU setup.

Have you managed to resolve this, or do you have any additional details? I'm also trying to use the QNN SDK with ONNX runtime on VOXL2 Mini.

Best,
Manu

@Eric-Katzfey Sure. Here is the link of the new post: https://forum.modalai.com/topic/3649/trouble-in-running-the-mavros_test-project-simple-example?_=1721124109041

@Eric-Katzfey I figured out the problem. I changed the mavlink pipe to mavlink_onboard keeping the channel number as 1. Setting the channel number as 2 seems to fix the problem. I haven't looked into the implementation of modal-pipe architecture but I thought that the channel number is just used by the client of the pipe (kind of like an open file descriptor number?). Basically it looks like having the channel number as 1 breaks the mavlink pipe somehow.

@enrico-bandera-0 It's not easy and, unfortunately, we cannot support that right now.

@tahawaru Okay, that shows you have 1.14.0-2.0.78. If you want to build 1.14.0-2.0.63 then you need to check that version out instead.

@Zachary-Lowell-0 as mentioned I have set offboard-mode in voxl-vision-hub.conf to "off".

I did a flight test shifting to offboard-mode,

The drone was able to execute the script in a stable way without any jittery motions or trying to come back to home position.

These are the flight logs,
https://review.px4.io/plot_app?log=e32a28df-cbba-4765-8499-6fe46e2ff890

@Eyucel To use PX4 HITL on VOXL 2 mini you need to connect to a DSP UART, not an applications processor UART. So, instead of using J10, you need to use a UART on J19 (Like the ESC UART pins 7 and 8).

@Alex-Kushleyev I see, that's unfortunate but thanks for letting me know.

@cegeyer

Here is a command to stream video from VOXL2 camera via ssh X forwading. FPS may be limited due to bandwidth (perhaps this is the command you meant in your reference above).

ssh -X root@<voxl-ip> gst-launch-1.0 qtiqmmfsrc camera=0 ! "video/x-raw, width=1280,height=720,framerate=30/1" ! autovideoconvert ! autovideosink

This should pop up a window with the stream from the camera. Make sure voxl-camera-server is not running if you are testing one of the mipi cameras connected to voxl2.

Alex

@Samuel-Lehman If this is a UVC camera you may be able to use voxl-uvc-server in order to create an MPA pipe and have tflite server read from that pipe

It may take some slight tweaking to get voxl-uvc-server to support your specific camera

@Samuel-Lehman , Not sure what is going on but i was able to build your example with the following command on VOXL2. I had to comment out #include "hello_cross.h" because i did not have it.

voxl2:~$ gcc -O2 -Wall voxl_io_uart_test.c -o voxl_io_uart_test -lmodal_pipe -lvoxl_io -L/usr/lib64 voxl_io_uart_test.c:14:12: warning: 'opt' defined but not used [-Wunused-variable] static int opt = 0; ^~~ voxl2:~$

I am not sure what's going on, you can enable verbose CMake file to see what compile command CMake is generating and compare that to the above.

Alex

@Judoor-0 Thanks for the contribution, this is very helpful!

@arcj , I don't think we have tried it but i believe the approach you are looking for is described here : https://developers.redhat.com/blog/2015/04/28/remote-debugging-with-gdb

If you are looking to use a GUI-based debugger front end, it would have to be compatible with gdb.

If you try this, let us know if it works! 🙂

Alex

Awesome, thanks!

@Samuel-Lehman I think the issue is with the tflite version running on voxl, which does not support v3 ops. An update to tflite on the voxl is being pushed but it will take some time to be merged into the main SDK. However, you can try tensorflow 2.17.1 from this branch.

Note that since it is not completely integrable, running it is slightly more complex. Make sure you are using the newest voxl-cross image to build voxl-tflite-server.

On Voxl, it cannot run directly, and you would have to run it via a Docker container. You can find a sample Dockerfile provided in docker_samples/ in the 'update_tf' branch. For further context on integrating opencl please refer to this.

Hopefully, the updated tflite will contain the necessary v3 op and it works smoothly.

@Aaron-Porter ,

Are you asking about a specific camera? (which one)

In general, you should not specify max exposure (in camera server config) which exceeds 1/fps. This can be often overlooked, so we recently added a check that enforces maximum exposure to be slightly less than 1/fps : https://gitlab.com/voxl-public/voxl-sdk/services/voxl-camera-server/-/commit/aa3ce587f6be769e1b07edf57f66e1eec40d4644

However, in some special cases, there may be other limitations. One case that I know of is the synced AR0144 at 30fps has a maximum exposure of less than half of 1/fps, otherwise it will skip frames, so our defaults for AR0144 are set to have 12ms maximum exposure. Also, you don't want to set the exposure too high for tracking cameras, since you will get too much motion blur.

Alex