30 September 2026

 

A box arrives at a warehouse workstation a few centimetres away from where it was expected. A camera can still read its label. A robot programmed to pick it up from a fixed position may miss it altogether.

To adjust, the robot needs to locate the box in three dimensions, work out how it is oriented and change its approach. That is the technology at the centre of Cognex’s agreement to acquire RealSense for approximately $500 million.

Cognex has spent more than four decades developing machine vision systems for factories and distribution centres. RealSense makes depth cameras and related technology used in robotics. Cognex says RealSense will expand its business in robotic perception: the systems that help machines judge where things are as they move and work around them. The deal is expected to close in the fourth quarter of 2026, subject to customary conditions. 

 

How do depth cameras help robots see in 3D?

 

An ordinary image records a scene in two dimensions. Software may recognise a box in that image, but recognition alone does not tell a robot precisely where to place its gripper. A depth camera adds estimated distances across the scene, helping the robot determine the position and shape of what it sees.

RealSense’s stereo depth cameras use two sensors looking at the same scene from slightly different positions. The system matches features in the two images and calculates their distance from how far those features appear to shift. Some models project an infrared pattern onto surfaces that otherwise offer little detail to match, such as a plain wall. Processing hardware turns the measurements into depth data that robot software can use. 

For the misplaced box, that data can help establish where its edges are and how it sits relative to the robot arm. Software must then choose a grip, plan a movement and control the arm as it reaches. If the box moves again, the system may have to reassess.

Important to note here is that a depth camera does not make a robot autonomous on its own. It supplies information for decisions and movements that must also work reliably. Yet without an accurate account of the surrounding space, even capable software can instruct a robot to reach for something that is no longer there.

The need extends beyond picking up boxes. A mobile warehouse robot must detect obstacles in its path. An arm handling mixed parts must find each one when they arrive in different positions. Underscoring that the technology can be used in diverse scenarios, RealSense says its technology is used in applications involving fixed robot arms, autonomous mobile robots, quadrupeds and humanoids. 

 

What Cognex and RealSense bring together

 

Cognex is already well acquainted with artificial intelligence and three-dimensional vision. Its products inspect goods, measure features, identify parts and read codes for manufacturers and logistics operators. It says it has more than 30,000 customers worldwide. RealSense adds a business focused on depth cameras for robots, including machines that navigate changing surroundings.

Cognex chief executive Matt Moschner says the combination could span industrial identification, 2D and 3D measurement, depth perception and robotic navigation. That describes the company’s intended direction; Cognex has yet to show how it will bring all those capabilities together in future products. RealSense chief executive Nadav Orbach says Cognex’s industrial customer base and sales reach would have taken his company years to build independently.

The combined company will have plenty of competition. In industrial machine vision, Cognex competes with companies including Keyence, Zebra Technologies and SICK. In depth cameras for robotics, RealSense faces suppliers such as Orbbec, Stereolabs and Luxonis. Their products differ in sensing methods, processing and software support. For robot makers, the choice also depends on how readily a camera fits into a machine and how dependably it works under the conditions that machine encounters. 

 

Why is Cognex buying RealSense now?

 

RealSense began within Intel in 2014 and was spun out in 2025. Cognex expects it to generate $80 million to $90 million in revenue in 2026, more than 50 per cent above the previous year. Its facial authentication product line is due to be spun out before the acquisition closes, leaving the business Cognex is buying focused more closely on robotic vision.

Cognex sees room for that business to grow. It estimates the robotic perception market at about $600 million today and projects growth of more than 25 percent a year to approximately $1.6 billion by 2030. 

RealSense already has cameras, customers and developers working in robotics. Buying the company gives Cognex a quicker route into that market, while its own industrial relationships could help RealSense reach more customers. 

For customers, the eventual test is practical. Can the combined company make robotic vision easier to integrate into machines that perform reliably when lighting, objects or surroundings change? RealSense strengthens Cognex at the sensing stage. It will take more than a sensor to make the entire robot adaptable.

A robot on a conventional production line may work brilliantly until a part arrives a few centimetres out of place. A more adaptable one could find the part, work out how it has moved and adjust its approach. That sounds like a small improvement. Across a factory or warehouse full of objects that never land in quite the same position twice, it could change which jobs are practical to automate.

 

Further reading on MoveTheNeedle.news:

• How Peridio and NXP are solving physical AI’s deployment problem — on the computing and software needed to keep intelligent devices working in the field. 
• Rebl Industries secures IKEA and H&M as customers for robots as a service — a look at deploying robots in changing commercial environments.
• The real robotics revolution is happening in the jobs people avoid — practical examples of robots at work in logistics.