SkillReal Signs Tier 1 Supplier for NVIDIA-Powered AI Inspection

SkillReal's system delivers sub-millimeter dimensional analysis of panels and parts using only off-the-shelf cameras, computers and its own algorithms.

May 12, 2025

Chris Clonts

An example of the Skillreal dashboard software that clients can see. It is updated live and integrates with product lifecycle management (PLM) systems. (SkillReal)

AI inspection company SkillReal has signed a contract to supply its system to a major Tier 1 global supplier in addition to global OEMs and other suppliers already using the system. The company shared the news in a call with SAE Media on the eve of Automate, the nation’s largest automation expo, which is this week in Detroit. SkillReal will show the system, which uses off-the-shelf camera and computing tech to detect defects far better than humans can and achieves an efficiency rate approaching 100%, at the expo.

The Skillreal system can be used at multiple points in any manufacturing process. (SkillReal)

SkillReal uses NVIDIA GPUs, Tensor RT acceleration, and compute unified device architecture (CUDA) programming to optimize and fuse large vision models with ultra-precise dimensional analysis to accurately inspect parts, bodies-in-white and assembly lines at a lower cost than traditional inspection. Photos are compared to a CAD digital twin.

The result, said SkillReal COO Peter Grabowski, is sub-millimeter accuracy in three dimensions for studs, MIG welds, weld nuts, spot weld points, holes and other features. The images come from everyday high-res webcams, he said. “Then we pair that with the NVIDIA GPUs to do the billions of calculations required to perform not only the presence check but the dimensionals also.”

Peter Grabowski in 2024 with the part of the tech that, combined with SkillReal's algorithms, can improve the speed and accuracy of parts inspections. (SAE | Chris Clonts)

He said humans are 80% accurate at inspection jobs. “Even your standard off-the-shelf AI is at 90 to 95%, and then we're taking it one step further at scale. It levels up to 99.7% with our proprietary algorithms,” he said. The system can look for 500 features in 15 seconds instead of 30 features in 45 seconds. Generally, that means assembly lines are never slowed or stopped to wait for inspections to catch up.

Being able to detect in multiple dimensions means that instead of merely checking to see if a feature is there (the goal of many inspectors), SkillReal can judge other important aspects. “Is the weld good? Is there a continuous bead? Are there no holes in it? Things like that,” Grabowski said.

The goal of any kind of inspection is to reduce assembly-line stoppages and expensive recalls, which create $30 billion-a-year across the industry in recalls and rejected parts that become scrap.

A new feature is that the former portable system is now built to be mounted at fixed spots along a line.

Grabowski said that data captured by the software is used to help continually train the SkillReal AI, but that the client keeps ownership of it. The fact that the company’s AI models are pre-trained means the system does not need to “learn” a new part with repeated operations. Once the camera is set up and the CAD is available, it’s ready to go.

He said the system is available now but is about 18 months away from wide commercial availability.

Ford's visual AI system

Skillreal is not the only company pursuing this sort of quality inspection. Ford, for instance, has deployed its Mobile Artificial Intelligence Vision System (MAIVS), a self-developed solution that is deployed at hundreds of stations in Ford plants worldwide. At the Kentucky Truck Plant in Louisville, the system uses iPhones. Its unique feature is that the iPhone is used as a self-contained unit, not only taking the reference photos but then processing the AI analysis on board before communicating via dashboard to a nearby operator. It’s the latest iteration of the system, which debuted in 2021.

MAIVS is often mounted on robots that allow it to look inside vehicles on the line before it generates a pass/fail status on the inspected part or parts.