@Vinny Perfect, thanks!

@Alex-Kushleyev Thank you so much!

I'm a bit preoccupied with a few other projects but hopefully I can take a look by the end of next week and get back to you.

@Eric-Katzfey I got about a million warnings when actually building but it did appear to build after initializing the submodules. Thanks!

Hi @Jetson-Nano
Sorry for my delay, I was OoO when the first message came through.
In any event, I can only say we have no plans at the moment to explore this new Lepton, and I have not even reviewed the spec sheet, so I cannot offer any guidance on it.
Sorry I can't be of more help.

picture with 130% voxl-camera-server useage

c98837c4-370d-4d40-88e7-7a57d0263137-image.png

voxl-camera-server.conf:

{
"version": 0.1,
"fsync_en": true,
"fsync_gpio": 109,
"cameras": [{
"type": "ar0144",
"name": "tracking_front",
"enabled": true,
"camera_id": 0,
"fps": 30,
"en_rotate": true,
"en_preview": true,
"preview_width": 1280,
"preview_height": 800,
"en_raw_preview": true,
"en_misp": true,
"misp_width": 1280,
"misp_height": 800,
"misp_venc_enable": true,
"misp_venc_mode": "h265",
"misp_venc_br_ctrl": "cbr",
"misp_venc_Qfixed": 38,
"misp_venc_Qmin": 15,
"misp_venc_Qmax": 50,
"misp_venc_nPframes": 9,
"misp_venc_mbps": 1.2,
"misp_awb": "auto",
"misp_gamma": 1,
"misp_zoom": 1,
"ae_mode": "lme_msv",
"gain_min": 54,
"gain_max": 8000,
"ae_desired_msv": 60,
"exposure_min_us": 20,
"exposure_max_us": 12000,
"exposure_soft_min_us": 5000,
"ae_filter_alpha": 0.6,
"ae_ignore_fraction": 0.2,
"ae_slope": 0.05,
"ae_exposure_period": 1,
"ae_gain_period": 1
}, {
"type": "imx412",
"name": "hires",
"enabled": true,
"camera_id": 1,
"fps": 30,
"en_rotate": false,
"en_preview": false,
"preview_width": 640,
"preview_height": 480,
"en_raw_preview": false,
"en_small_video": true,
"small_video_width": 1024,
"small_video_height": 768,
"small_venc_mode": "h264",
"small_venc_br_ctrl": "cqp",
"small_venc_Qfixed": 30,
"small_venc_Qmin": 15,
"small_venc_Qmax": 40,
"small_venc_nPframes": 9,
"small_venc_mbps": 2,
"en_large_video": true,
"large_video_width": 4056,
"large_video_height": 3040,
"large_venc_mode": "h264",
"large_venc_br_ctrl": "cqp",
"large_venc_Qfixed": 40,
"large_venc_Qmin": 15,
"large_venc_Qmax": 50,
"large_venc_nPframes": 29,
"large_venc_mbps": 40,
"en_snapshot": true,
"en_snapshot_width": 4056,
"en_snapshot_height": 3040,
"exif_focal_length": 3.1,
"exif_focal_length_in_35mm_format": 17,
"exif_fnumber": 1.24,
"snapshot_jpeg_quality": 75,
"ae_mode": "isp",
"gain_min": 54,
"gain_max": 8000
}, {
"type": "ar0144",
"name": "tracking_rear",
"enabled": true,
"camera_id": 2,
"fps": 30,
"en_rotate": true,
"en_preview": true,
"preview_width": 1280,
"preview_height": 800,
"en_raw_preview": true,
"en_misp": true,
"misp_width": 1280,
"misp_height": 800,
"misp_venc_enable": true,
"misp_venc_mode": "h265",
"misp_venc_br_ctrl": "cbr",
"misp_venc_Qfixed": 38,
"misp_venc_Qmin": 15,
"misp_venc_Qmax": 50,
"misp_venc_nPframes": 9,
"misp_venc_mbps": 1.2,
"misp_awb": "auto",
"misp_gamma": 1,
"misp_zoom": 1,
"ae_mode": "lme_msv",
"gain_min": 54,
"gain_max": 8000,
"ae_desired_msv": 60,
"exposure_min_us": 20,
"exposure_max_us": 12000,
"exposure_soft_min_us": 5000,
"ae_filter_alpha": 0.6,
"ae_ignore_fraction": 0.2,
"ae_slope": 0.05,
"ae_exposure_period": 1,
"ae_gain_period": 1
}, {
"type": "pmd-tof-liow2",
"name": "tof",
"enabled": true,
"camera_id": 3,
"fps": 10,
"en_rotate": true,
"ae_mode": "off",
"gain_min": 0,
"gain_max": 0,
"exposure_max_us": 6000,
"standby_enabled": false,
"decimator": 5
}, {
"type": "ar0144",
"name": "tracking_down",
"enabled": true,
"camera_id": 4,
"fps": 30,
"en_rotate": false,
"en_preview": true,
"preview_width": 1280,
"preview_height": 800,
"en_raw_preview": true,
"en_misp": true,
"misp_width": 1280,
"misp_height": 800,
"misp_venc_enable": true,
"misp_venc_mode": "h265",
"misp_venc_br_ctrl": "cbr",
"misp_venc_Qfixed": 38,
"misp_venc_Qmin": 15,
"misp_venc_Qmax": 50,
"misp_venc_nPframes": 9,
"misp_venc_mbps": 1.2,
"misp_awb": "auto",
"misp_gamma": 1,
"misp_zoom": 1,
"ae_mode": "lme_msv",
"gain_min": 54,
"gain_max": 8000,
"ae_desired_msv": 60,
"exposure_min_us": 20,
"exposure_max_us": 12000,
"exposure_soft_min_us": 5000,
"ae_filter_alpha": 0.6,
"ae_ignore_fraction": 0.2,
"ae_slope": 0.05,
"ae_exposure_period": 1,
"ae_gain_period": 1
}]
}

Hey @svempati
I trained a model in the docker command as provided by you. I am able to run an inference on potholes following exactly the steps provided by @Zachary-Lowell-0 earlier.
Could I ask you what your voxl-tflite-version is? And if you havent yet, could you try to run this version: http://voxl-packages.modalai.com/dists/qrb5165/sdk-1.6/binary-arm64/voxl-tflite-server_0.5.1_arm64.deb
Make sure to change the .tflite model and labels as shown in the loom video after the install, then run voxl-configure-tflite

Thanks

Hello @Cliff-Wong,

The figure 8 no longer works. It is now crashing into the netting in the room after starting the figure 8 mode. Though, it did in the past.

Here is the screen recording as it was happening Link

I also notice that it seems a sensor on the bottom has come loose due to the crashes sustained while trying to test the navigation. As well as the battery retention clips has broken off. Link

@emarcphera @Eric-Katzfey thank you so much after upgrading to voxl-mainline im getting GPS readingScreenshot from 2026-02-28 01-45-10.png

@Sarika-Sharma We have a variety of parameters sets available in the voxl-px4-params repo https://gitlab.com/voxl-public/voxl-sdk/utilities/voxl-px4-params/ However, they are mostly for either indoor or outdoor use. Older versions of PX4, like what our fork is based off of, had a hard time with external vision based navigation coexisting with GPS based navigation. Newer versions of PX4, supposedly, are much better at this but we have not spent the time to prepare a parameter set for that yet. You are free to experiment with our mainline PX4 build here: http://voxl-packages.modalai.com/dists/qrb5165/dev/binary-arm64/voxl-mainline-px4_1.17.0-1.0.2-202602240859_arm64.deb. However, if you decide to use the mainline build, be aware that it is considered "experimental" at this point and has very little testing on voxl2 so use extreme caution.

@mark , it looks like your version of voxl sdk has an early version of the imx664 driver (not supporting 2704x1540 resolution.

Are you able to upgrade sdk to 1.5.x or 1.6.x?

Alternatively, i could share the latest camera drivers to use with 1.4.5.

I would also recommend updating voxl-camera-server to the latest, but that would also require updating other packages.. would you consider upgrading the sdk or are you locked to 1.4.5?

Alex

@voxltester,
When you say voxl-microdds-agent doesnt work, do you mean it doesnt run? Or it doesnt create the topics? Whats the terminal output when you run it directly from the terminal?
Is your setup like earlier where you have the hat on and the wifi dongle is connected on the hat?

If the topics are only showing up after disabling the wifi interface, then I suspect its a ip routing issue, and that voxl microdds agent is binding to the first interface that comes up, which is the wifi.

@jh81306
Hello there,
Yes, real time performance is achievable with 3 cameras at 30fps, evidencing the benefits of our hardware accelerated GPU feature tracker. From this perspective, if you plan on using GPU-heavy applications, QVIO may be the second best option, as it uses remote procedure calls between the linux processor and the DSP, freeing GPU usage. However, due to the inherent nature of the DSP (L2 Cache, etc) and timing considerations, running QVIO on 3 Cameras may not guarantee a consistent update rate depending on the processor load/scheduler priorities. In essence, choosing one or the other is a trade-off that should be based on user use-case.

In terms of estimator robustness, our branch of OpenVINS has been considerably modified to support accurate estimates under degraded conditions, and all ModalAI UAV platforms have gone through extensive calibration in order to obtain the optimal parameter set. Hence, in terms of estimator quality, OVINS outperforms QVIO, but requires carefully calibration of your visual-inertial sensors. Finally, If you opt to acquire a ModalAI Vision drone, OpenVINS should be fully functional out-of-the-box, facilitating a potential pipeline transition.

@Leo-Allesch , if you are talking about VSLAM, which stands for Visual Simultaneous Localization and Mapping, then this is a sub-class of SLAM which uses only cameras (and often IMUs, then it is VISLAM).

VSLAM and VISLAM usually generate and maintain a sparse map of 3D features (points), which is used for self localization.

If you want to be able to detect a target, and you have your 3D pose from VIO (VIO = VISLAM), then you can either triangulate the target location or get it's location from a single frame if you know the size of the target (such as April tag, etc). Then you can transform the coordinate of the tag from the camera frame into your "world" (VIO) coordinates.

The depth sensor would allow you to build a more detailed 3D map / point cloud using the 3D pose output from the VIO algorithm (or just using a depth sensor, which may be more complicated). You can also build a dense 3D map using camera only (and IMU), but that typically requires a lot more computation than using a depth sensor.

You don't necessarily need a depth sensor for identifying the position of a specific target that is easily detected by a camera.

There is a lot of research material you can find on these topics, please look into it deeper and let us know if you have specific questions about using VOXL2 hardware and software.

Alex

@Catalystmachine , for questions regarding orders, please reach out here : https://www.modalai.com/pages/contact-us

Hi @Hunter-Scott ,

We are looking into this. We may not have an exact CAD model for every single variant of the products. Is there something specific in your product that significantly deviates from the provided CAD files? Maybe we can provide something that is close (but not exactly correct).

Alex

Hi @austin-c
yeah, and it will not.
There are dozens of changes between the two boards.
It's just easier and cheaper to order the correct version. We have no documents showing how to convert it is simply not doable for the cost and risk and time.

Not the news you wanted, but please trust me on that one. 🙂

@Moderator

Thanks for the clarification on the ~50 ms glass-to-glass number.

Trying to get a better understanding the performance envelope:
1. Does the 50 ms include RF transport over Doodle, decode, and HDMI output latency, or is that measured only within the VOXL + VRX software stack?
2. Was this measured in point-to-point mode or mesh routing mode?
3. Can you expand on bitrate, resolution, and encoder settings were used (GOP structure, B-frames disabled, UDP vs RTSP)?
4. Is MAVLink or other traffic running on the same link during that measurement?
5. Do you have a latency breakdown per stage (encode / RF / decode / display)?

Appreciate any information you can provide. I just want to make sure I can test and validate this effectively.