@jimbow77 those drones are for aeronautical research. we recommend installing a dronetag or similar

@ggangmi What was the flight mode on takeoff and what did it unexpectedly change to? There should be logs in /data/px4/log. They may be incomplete or maybe even corrupted if the drone lost power in the crash but there should be something there. The logging starts when the drone is armed.

@Sarika-Sharma , Since you are testing different configurations, i would suggest to reduce the number of cameras first to find working cases and see if you may have a bad camera or a bad flex cable. So, try both tracking camera one at a time, then hires camera by itself, then two tracking cameras, then two tracking + hires, etc.. Alex

@Jetson-Nano , We have not tried QVIO with Boson, it may work (but we can't really support that). The boson output is usually a bit fuzzy, so i am not sure how the feature detection will work, but it may be ok. Disabling FFC will definitely resolve the gaps in the frame feed, but disabling the FFC will also affect the Boson image quality, which may degrade over time. If you wanted to use any USB (thermal) camera with VIO, you would just have to use voxl-uvc-server to publish the images to mpa, so that voxl-qvio-server can subscribe to them. The only issue is that the uvc interface does not provide accurate timestamp of the camera, but it may be ok just to use the time of arrival of the frame. QVIO does estimate the time offset between the IMU and camera, so as long as the timestamp is consistently captured, it should be ok.. Alex

Hi @Richa We have not provided our pinouts for this and have no plans to. For our Micro Coax direct CMOS based sensors, you need to use our products end-to-end to guarantee functionality. Hope this helps clarify.

@Colby-Howell You can save camera data onboard using voxl-logger, but it will take up a lot of storage very quickly and the voxl2 has < 64gb so be careful with that. If you're not trying to save raw frames and are okay with compression & artifacts you can stream it to a ground computer with voxl-streamer and save it there For connecting to the drone the easiest (especially cause you don't have a ground radio) is wifi, you can just connect your drone to a somewhat powerful router, set a static IP and ssh in. For autonomous navigation, the best way in my opinion is to talk to voxl-px4 through the uxrce-dds bridge (implemented on the voxl2 as voxl-microdds-agent) https://docs.px4.io/main/en/middleware/uxrce_dds

@Piyush-Singh , The voxl-portal does not handle well streaming large images (including in multi-view, where the images are actually appearing small). The voxl-portal back-end uses software JPEG encoding, so one of your streams (hires_large) is 4208x3120, which takes LOT of cpu processing to encode to jpeg. You can see that the voxl-portal process uses 260% cpu (2.6 full cores), which is a lot. There is no practical need for streaming small and large uncompressed images to voxl-portal, so you can just enable small video stream, which is much smaller. Also, if you select an individual camera stream, you can enable advanced camera controls (click on the small check box in lower left) and then you can adjust the JPEG encode quality from default 70 to something much lower in order to use less cpu. This setting will be applied to all cameras and you can go back to multi view after that. Note that the jpeg encode quality only affects voxl-portal view, not h264/h265 encoding. Please try. Alex

I'm losing video feed flying just behind a wall not far from me. What can I do to prevent video signal loss? Increase transmit watts or is there a frequency better suited to penetrate thick concrete walls?

@Alex-Kushleyev Could you please share the pinouts of the serial console port. I am not able to find it anywhere.

Hi @Peter-Lingås-0, I have been working on voxl-tflite-server for the last few weeks and was actually testing this. I find that both YOLOV8 and YOLOV11 work well with GPU delegate. The comment you are referring to is removed as of version 0.51 of voxl-tflite-server.

Hi @jon , OK got it. Actually, we already do have a passthrough mode that can accept raw data coming from linux (via voxl io server) and forward it to the UART connection for voxl2_io board, for reference, code is here : https://github.com/modalai/px4-firmware/blob/voxl-dev/src/drivers/voxl2_io/voxl2_io.cpp#L270 Let me figure out (remember) how to use it, I will update you soon. Alex

@paul-foley , it looks like our voxl-lepton-server is not setting the FPS field in the metadata : https://gitlab.com/voxl-public/voxl-sdk/services/voxl-lepton-server/-/blob/master/src/publisher.c?ref_type=heads#L338 . I do see above that we do set the framerate in json info about the output pipe (cJSON_AddNumberToObject(json, "framerate", 9);) and the framerate is hardcoded to 9, since i guess all lepton sensors are running at 9fps. I just made the change to publish the fps : https://gitlab.com/voxl-public/voxl-sdk/services/voxl-lepton-server/-/commit/110b9cce4c021322566e46f36d6878fca55f7d26 You are using an older voxl SDK, so installing latest nightly package (from http://voxl-packages.modalai.com/dists/qrb5165/dev/binary-arm64/ will probably not work). So you probably need to take the source of the version you are running now (1.3.3 : https://gitlab.com/voxl-public/voxl-sdk/services/voxl-lepton-server/-/tree/v1.3.3?ref_type=tags) and apply the same change in the source code and build it. Since i just pushed to dev branch, a new version of leptop server should appear in our package repo tomorrow, and you can download and try it. it might work, but your SDK is 2 major versions down from where we are now (1.4.x vs 1.6.x), so you may want to just build the patched version 1.3.3 : git clone https://gitlab.com/voxl-public/voxl-sdk/services/voxl-lepton-server.git cd voxl-lepton-server git checkout v1.3.3 #make the change in publisher.c to include fps ... ... #start voxl-cross docker (you can download it in our downloads section: https://developer.modalai.com/asset) docker run -it --rm -v `pwd`:/opt/code -w /opt/code voxl-cross:V4.6 bash ./install_build_deps.sh qrb5165 sdk-1.4 ./build.sh qrb5165 ./make_package.sh #then, outside docker, deploy the new package to voxl2 via adb ./deploy_to_voxl.sh Please let me know if this works for you! Alex

@Saleh-Ibrahim the drones come pre-configured for VIO and are flight tested in VIO mode prior to shipping. To troubleshoot VIO, please start here: https://docs.modalai.com/voxl-open-vins-server/#troubleshooting VOA does not ship enabled by default

@bendraper I just tried it and it is working now: modalai@modalai:~/development/modalai/voxl-vision-hub (dev)$ voxl-docker -i voxl-cross launching image: voxl-cross with the following command: docker run --rm -it --net=host --privileged -w /home/modalai --volume=/dev/bus/usb:/dev/bus/usb -e LOCAL_USER_ID=0 -e LOCAL_USER_NAME=root -e LOCAL_GID=0 -v /home/modalai/development/modalai/voxl-vision-hub:/home/root:rw -w /home/root voxl-cross /bin/bash -l voxl-cross(4.4):~(dev)(1.9.21)$ ./clean.sh voxl-cross(4.4):~(dev)(1.9.21)$ ./install_build_deps.sh qrb5165 dev using qrb5165 dev debian repo Ign:1 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ InRelease Ign:2 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ Release Get:3 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ Packages [76.8 kB] Fetched 76.8 kB in 1s (122 kB/s) Reading package lists... Done INSTALLING: libmodal-json libmodal-pipe librc-math libmodal-cv voxl-mpa-tools voxl-mavlink Reading package lists... Done Building dependency tree... Done Reading state information... Done The following NEW packages will be installed: libmodal-cv:arm64 libmodal-json:arm64 libmodal-pipe:arm64 librc-math:arm64 voxl-mavlink:arm64 voxl-mpa-tools:arm64 0 upgraded, 6 newly installed, 0 to remove and 2 not upgraded. Need to get 5026 kB of archives. After this operation, 0 B of additional disk space will be used. Get:1 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ librc-math 1.4.6-202510091209 [180 kB] Get:2 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ libmodal-cv 0.6.0-202510141028 [3887 kB] Get:3 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ libmodal-json 0.4.8-202510091202 [52.5 kB] Get:4 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ libmodal-pipe 2.14.10-202601210836 [151 kB] Get:5 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ voxl-mavlink 0.1.6-202511141246 [650 kB] Get:6 http://voxl-packages.modalai.com ./dists/qrb5165/dev/binary-arm64/ voxl-mpa-tools 1.5.6-202601201423 [106 kB] Fetched 5026 kB in 1s (6805 kB/s) debconf: unable to initialize frontend: Dialog debconf: (No usable dialog-like program is installed, so the dialog based frontend cannot be used. at /usr/share/perl5/Debconf/FrontEnd/Dialog.pm line 79, <STDIN> line 6.) debconf: falling back to frontend: Readline Selecting previously unselected package librc-math:arm64. (Reading database ... 35164 files and directories currently installed.) Preparing to unpack .../0-librc-math_1.4.6-202510091209_arm64.deb ... Unpacking librc-math:arm64 (1.4.6-202510091209) ... Selecting previously unselected package libmodal-cv:arm64. Preparing to unpack .../1-libmodal-cv_0.6.0-202510141028_arm64.deb ... Unpacking libmodal-cv:arm64 (0.6.0-202510141028) ... Selecting previously unselected package libmodal-json:arm64. Preparing to unpack .../2-libmodal-json_0.4.8-202510091202_arm64.deb ... Unpacking libmodal-json:arm64 (0.4.8-202510091202) ... Selecting previously unselected package libmodal-pipe:arm64. Preparing to unpack .../3-libmodal-pipe_2.14.10-202601210836_arm64.deb ... Unpacking libmodal-pipe:arm64 (2.14.10-202601210836) ... Selecting previously unselected package voxl-mavlink:arm64. Preparing to unpack .../4-voxl-mavlink_0.1.6-202511141246_arm64.deb ... Unpacking voxl-mavlink:arm64 (0.1.6-202511141246) ... Selecting previously unselected package voxl-mpa-tools:arm64. Preparing to unpack .../5-voxl-mpa-tools_1.5.6-202601201423_arm64.deb ... Unpacking voxl-mpa-tools:arm64 (1.5.6-202601201423) ... Setting up voxl-mavlink:arm64 (0.1.6-202511141246) ... Setting up libmodal-json:arm64 (0.4.8-202510091202) ... Setting up librc-math:arm64 (1.4.6-202510091209) ... Setting up libmodal-pipe:arm64 (2.14.10-202601210836) ... Setting up libmodal-cv:arm64 (0.6.0-202510141028) ... Setting up voxl-mpa-tools:arm64 (1.5.6-202601201423) ... Processing triggers for libc-bin (2.39-0ubuntu8.5) ... Done installing dependencies voxl-cross(4.4):~(dev)(1.9.21)$ ./build.sh qrb5165 Found voxl-cross version: 4.4 -- --------------------------------------------------------- -- Using voxl-cross 64-bit toolchain for QRB5165 ubun1 18.04 -- C Compiler : /usr/bin/aarch64-linux-gnu-gcc-7 -- C++ Compiler: /usr/bin/aarch64-linux-gnu-g++-7 -- Sysroot : /opt/sysroots/qrb5165_1 -- C flags : -isystem=/usr/lib/gcc/aarch64-linux-gnu/7/include -isystem=/usr/local/include -isystem=/usr/lib/gcc/aarch64-linux-gnu/7/include-fixed -isystem=/usr/include/aarch64-linux-gnu -isystem=/usr/include -idirafter /usr/include -march=armv8.2-a -- CXX flags : -isystem=/usr/include/c++/7 -isystem=/usr/include/aarch64-linux-gnu/c++/7 -isystem=/usr/include/c++/7/backward -isystem=/usr/lib/gcc/aarch64-linux-gnu/7/include -isystem=/usr/local/include -isystem=/usr/lib/gcc/aarch64-linux-gnu/7/include-fixed -isystem=/usr/include/aarch64-linux-gnu -isystem=/usr/include -idirafter /usr/include -march=armv8.2-a -- Link Flags : -B/opt/sysroots/qrb5165_1/usr/lib/ -L/opt/sysroots/qrb5165_1/usr/lib/gcc/aarch64-linux-gnu/7 -L/opt/sysroots/qrb5165_1/usr/lib/aarch64-linux-gnu -L/opt/sysroots/qrb5165_1/usr/lib -L/opt/sysroots/qrb5165_1/lib/aarch64-linux-gnu -L/opt/sysroots/qrb5165_1/usr/lib/aarch64-linux-gnu -L/opt/sysroots/qrb5165_1/lib -L/usr/lib64 -L/usr/lib -- The C compiler identification is GNU 7.3.0 -- The CXX compiler identification is GNU 7.3.0 -- Detecting C compiler ABI info -- Detecting C compiler ABI info - done -- Check for working C compiler: /usr/bin/aarch64-linux-gnu-gcc-7 - skipped -- Detecting C compile features -- Detecting C compile features - done -- Detecting CXX compiler ABI info -- Detecting CXX compiler ABI info - done -- Check for working CXX compiler: /usr/bin/aarch64-linux-gnu-g++-7 - skipped -- Detecting CXX compile features -- Detecting CXX compile features - done -- Configuring done (0.3s) -- Generating done (0.0s) -- Build files have been written to: /home/root/build [ 6%] Building C object utils/CMakeFiles/voxl-inspect-vfc.dir/voxl-inspect-vfc.c.o [ 6%] Building C object utils/CMakeFiles/voxl-calibrate-px4-horizon.dir/voxl-calibrate-px4-horizon.c.o [ 12%] Building C object src/CMakeFiles/voxl-vision-hub.dir/control_input.c.o [ 12%] Building C object src/CMakeFiles/voxl-vision-hub.dir/fixed_pose_input.c.o [ 18%] Building C object src/CMakeFiles/voxl-vision-hub.dir/geometry.c.o [ 18%] Building C object src/CMakeFiles/voxl-vision-hub.dir/autopilot_monitor.c.o [ 24%] Building C object src/CMakeFiles/voxl-vision-hub.dir/config_file.c.o [ 24%] Building C object src/CMakeFiles/voxl-vision-hub.dir/horizon_cal.c.o [ 30%] Building C object src/CMakeFiles/voxl-vision-hub.dir/main.c.o [ 36%] Building C object src/CMakeFiles/voxl-vision-hub.dir/mavlink_io.c.o [ 36%] Building C object src/CMakeFiles/voxl-vision-hub.dir/imu_manager.c.o [ 27%] Building C object src/CMakeFiles/voxl-vision-hub.dir/horizon_cal_file.c.o [ 39%] Building C object src/CMakeFiles/voxl-vision-hub.dir/mavlink_for_ros.c.o [ 42%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_backtrack.c.o [ 45%] Building C object src/CMakeFiles/voxl-vision-hub.dir/obs_pc_filter.c.o [ 48%] Building C object src/CMakeFiles/voxl-vision-hub.dir/misc.c.o [ 51%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_figure_eight.c.o [ 54%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_follow_tag.c.o [ 57%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_mode.c.o [ 60%] Linking C executable voxl-calibrate-px4-horizon [ 60%] Built target voxl-calibrate-px4-horizon [ 63%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_trajectory.c.o [ 66%] Linking C executable voxl-inspect-vfc [ 66%] Built target voxl-inspect-vfc [ 69%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_vfc.c.o [ 72%] Building C object src/CMakeFiles/voxl-vision-hub.dir/offboard_wps.c.o [ 75%] Building C object src/CMakeFiles/voxl-vision-hub.dir/pose_filter.c.o [ 78%] Building C object src/CMakeFiles/voxl-vision-hub.dir/state_manager.c.o [ 81%] Building C object src/CMakeFiles/voxl-vision-hub.dir/tag_manager.c.o [ 84%] Building CXX object src/CMakeFiles/voxl-vision-hub.dir/trajectory_monitor.cpp.o [ 87%] Building C object src/CMakeFiles/voxl-vision-hub.dir/transform_ringbuf.c.o [ 90%] Building C object src/CMakeFiles/voxl-vision-hub.dir/vio_manager.c.o [ 93%] Building C object src/CMakeFiles/voxl-vision-hub.dir/voa_manager.c.o [ 96%] Building C object src/CMakeFiles/voxl-vision-hub.dir/voxl_vfc_config.c.o [100%] Linking CXX executable voxl-vision-hub [100%] Built target voxl-vision-hub

Hello @HLaGrone , Sorry for the delay in response! Did something happen to the board that caused it to stop booting? Are you able to remove the voxl2 board completely from the drone and just try to boot it by itself (nothing plugged in except APM / ESC for power and USB)? Sometimes issues on power rails going to external components (like cameras) can prevent the board from booting. Alex

Hi @shawn_ricardo Yes, any I2C port on an expansion board has 2.2K pullups to the voltage on Pin 1 of that connector. The Linux Kernel user guide has info on which QUPS are enabled or not in which system image. [image: 1769099579133-3356687b-6e2d-4443-9a15-6e9aa07566d1-image.png] and specifically this helps: https://docs.modalai.com/voxl2-qups/ Note not all QUPs have actually been enabled in the system, but with the sets of ports we have enabled, we never had any customer claim of a shortage: