Today Luxonis has entrusted us with sharing two announcements with the community.
The first is that Luxonis has strengthened its commitment to the ecosystem by increasing their OSRA membership from Supporter to Silver. Their enhanced membership matches the company’s growth and success, and is evidence of the importance of robotics to Luxonis.
“We believe robotics should be accessible to anyone with an idea and the drive to build it. OSRA shares our commitment to open source and collaboration, and we’re proud to join that effort.”
–Max McLaughlin, Director of Product & Growth, Luxonis
The second is the release of the Luxonis OAK 4 FFC vision computer! Place up to five cameras anywhere on your robot and the tiny (70mm x 53mm) OAK 4 FFC handles sync, recalibration, depth, and AI between them all. It’s just released today and you can learn all the details at the Luxonis site, as well as check out an open source reference design for it on GitHub.
To celebrate these two happy announcements, Luxonis is offering a free OAK 4 FFC developer kit for a member of the Open Robotics community. The kit—which comes with the board and sensors—will sell for $849 USD and will ship them in November. To enter to win the kit, all you need to do is comment on this post and tell Luxonis what you’d do with it. You have a week (until November 3rd), after which the Luxonis team will choose their favorite comment. That author will win the kit!
We’re grateful to Luxonis for the continuing support and look forward to working with them even more closely to increase the open source robotics ecosystem.
This looks amazing, I wish I get this !!! I’m currently using the OAK-D pro wide on a custom built drone to do site inspections in low light as a part of my PhD research, but we are struggling with flight time since we are using the RPi with a hailo accelerator to do our computing and my setup is a bit bulky. I would love to use something like your dev kit which looks fairly lightweight and would help me shave some grams of my drone while also getting access to a multi view of the surroundings. My current setup only has a front facing camera which I pass through my custom vSLAM algorithm to enable better tracking and mapping in low light.
Super exciting! If we win your dev kit we would use it in our free-flying robot called Orb launching to the International Space Station in May 2027. We already use your Luxonis OAK-D Pro W camera in prototyping, and individual modules could make it easier for us to have a full collision avoidance system and localization on all sides of the robot. A unique problem when you can float in any direction in zero-gravity
I’m currently building out an XLeRobot (Prince on X: "New robot setup coming soon by @tnkrdotai" / X) from makermods and tnkr.ai which is bimanual mobile manipulator (dual SO-101 arms on a holonomic base) and designing a custom 3D-printed 2-DOF anthropomorphic head for immersive VR teleoperation with Meta Quest 3
Right now, getting low-latency stereoscopic vision for the VR headset plus dual wrist cameras and base perception creates a massive USB bandwidth and sync headache. The OAK 4 FFC solves this on a single board:
Human-IPD Active Stereo in the Head: I’ll mount a stereo pair in the pan-tilt humanoid head spaced precisely at human interpupillary distance (~63–65mm) with hardware frame sync. As the operator turns their head in VR, the robot head tracks the IMU pose in real time while streaming synchronized stereo RGB-D back to the headset for zero-nausea depth perception during fine bimanual manipulation.
Dynamic Calibration on 3D-Printed Mounts: Because 3D-printed neck mechanisms flex and shift over time, Luxonis Dynamic Calibration on RVC4 will keep the custom stereo baseline aligned using scene texture without needing a checkerboard recalibration before every teleop session.
5-Camera Unified Pipeline for LeRobot VLA Training: With 5 FFC slots on one board, I can run the 2 stereo “eyes” in the head alongside a chest/base navigation camera and gripper overview cameras—offloading depth and object tracking to the 52 TOPS on-device compute so the main compute node stays free for ROS 2 control and recording synchronized imitation learning datasets.
I plan to open-source the CAD files for the OAK 4 FFC humanoid head mount alongside the ROS 2 / LeRobot VR teleop pipeline so the whole community can drop it onto their builds!
I’d build an OR (Operating Room) compliance monitor, basically a second set of eyes for the surgical team that never gets tired.
The problem: a lot of surgical site infections and “never events” come from small protocol slips, not big mistakes. Someone reaches over the sterile field, the door opens 40+ times during a hip replacement, a sponge isn’t accounted for at closing, the time-out gets rushed. Today most of this is checked manually or not at all, and nobody wants a camera streaming OR footage to a cloud server for obvious privacy reasons.
The idea: mount the FFC modules around the room (one over the table, one on the door, one on the scrub/prep area, one on the instrument tray) and run everything on-device:
Sterile field tracking: flag when a non-scrubbed person gets too close to the sterile zone, using depth instead of just 2D boxes
Door traffic counting, since every opening raises infection risk
Sponge and instrument counts at the tray, cross-checked before closing
Checking that the surgical safety checklist / time-out actually happened before the first incision
Quiet alerts to the circulating nurse instead of loud alarms that interrupt the surgeon
Why the OAK 4 FFC: this project basically doesn’t work without it. Hospitals won’t accept raw video leaving the room, and with the processing on the device itself, only anonymized events go out (like “door opened at 10:42” or “sterile zone breach, 2s”), never footage. The FFC form factor means I can put cameras exactly where the angles are needed from a single compute board instead of building four separate systems, and the depth makes the “too close to the sterile field” check reliable, which plain RGB can’t do well. Hooking it into ROS 2 also makes it easy to log events and later plug into a robotic instrument-handling arm.
I’d start with a mock OR setup and then pilot it with a surgical team to see which alerts actually help versus just add noise. I’m already Working with Luxonis Client Projects using OAK4 D Pro Camera on Special Patient Zone SPZ GuardianWHO 5 Hand Moments Compliance Monitor(See Demo) in Hospital Wards. Would love to share the results back with the community .
I suppose I’ll toss my hat in the ring. I’ve been developing an affordable autonomous RC car system based on ESP32 and Raspberry Pi. The system currently uses an OakD Lite camera for basic object detection currently. A higher power board and camera sensors would allow for a much more capable SLAM system and Autonomy features. It would bring the goal for the project out of the scope of $500 all in cost. SauRCon Code
It’s always really refreshing to see companies integrate super closely to the open source movement (though Luxonis IMO has always been great with that). Congrats to Luxonis on the launch of their product!
If I were to get access to an Oak 4 FFC kit, I’d probably give it my University group. I help run a summer AI-STEM camp at UMD’s Southern Maryland satellite campus (USMSM). It’s a pretty rural area, and most of our ~50 high schoolers (rising 9th-12th graders) don’t get anything like robotics or autonomy in their classes.
The camp is built around the (soon to be entirely open source) Mini Claw Rover, a small rover we designed in-house. In the last week, students come up with their own projects on it and go after whatever problem they want, with us there to help. Right now the rover sees through an ESP32 camera and has to stream to a laptop or the cloud for anything heavy. A few things we and the kids have gotten working that way:
:
:
And :
We looked at a Jetson Orin Nano dev kit, but it’s just too big for the rover (103 x 90.5 mm). The OAK 4 FFC at 70 x 53 mm actually fits, so everything could run on the rover itself.
That would give our older students a real step up. They could try 3D mapping with depth from multiple cameras, or run a small open model like SmolVLM right on the camera and let the rover decide what to do based on what it sees. It’s a fun, hands-on way to show them the stuff research labs are working on, and they’d have the freedom to experiment with it while we guide them.
.
