Luxonis is an American company that created the OAK (OpenCV AI Kit) camera family: depth cameras with a built-in processor for neural networks. Each camera we know from datasheets has an Intel Movidius Myriad X vision chip that computes stereo depth and runs AI models on the device itself. The robot or main computer therefore receives a ready depth image or detection result rather than a raw pixel stream. This approach is called edge processing.
The line is broad. The OAK-D is the classic camera with three sensors: two greyscale stereo cameras and a central colour camera. The OAK-D Lite is a small, cheaper version measuring 91×28×17.5 mm, with a 13 MP colour sensor and USB-C power; the maker stresses that it runs the same models as the older OAK-D because it has the same processor. The OAK-D Pro adds active illumination: an infrared projector casting about 4,700 laser dots, and a flood LED for low light, which helps on uniform surfaces. PoE versions such as the OAK-D S2 PoE have an IP67 housing, gigabit Ethernet with power over the cable and an M12 connector, so they suit halls and mobile robots. The OAK-D CM4 PoE has a Raspberry Pi CM4 module built in and can work stand-alone, accessible over SSH. OAK-SoM modules are sold separately for embedding in a custom board. The datasheets list industrial automation, robotics, surveillance and security systems among applications.
In which robots? No robot in our catalogue has a Luxonis camera in its specification, so we will not point to a model. This family goes rather into robots built by integrators, universities and start-ups than into ready-made humanoids. Typical places are mobile platforms, such as those in mobile robots, and prototypes in laboratories and schools. Its counterparts in finished robots are Intel RealSense cameras with a separate vision processor and Orbbec, while heavy AI models are rather run on NVIDIA Jetson.
For the cost and performance of a robot, a camera with an AI processor relieves the main computer. The price is the limited power of the chip itself: the Myriad X handles light networks for object detection and depth, but not the large language models or VLA models described in “VLA Models: How AI Controls Humanoid Robots in 2026”. For a warehouse robot, as in “AMR Robots in the Warehouse”, it means a cheap, compact obstacle-avoidance sensor, and for a humanoid an addition rather than the main perception system. The safety scanner required by standards is replaced by other devices, for example from SICK; an AI camera will not substitute for it.
For Poland and Europe the materials name no distributor. The cameras are available from the maker's online shop, and some Polish firms, such as Antmicro, work on Edge AI systems and vision modules for robotics, although our materials contain no evidence that it offers Luxonis in particular. Note that the datasheets are dated December 2021, so current versions and availability must be checked before purchase. Camera data also raises privacy questions, covered in “Data Security: Robots with Cameras and the GDPR”.
Products
- OAK cameras with integrated edge AI (depth + vision) for robot perception
Sample specs
OAK-D CM4 PoE: 3 integrated cameras, Myriad X VPU for on-device depth computation, PoE power and connectivity via the built-in RPi CM4 module (source: OAK-D-CM4-POE_Datasheet.pdf).
Relevance for integrators in Poland
Market overview for Polish integrators; the company does not build its own finished robots.