Widespread use of industrial physical AI

"Thanks to Nvidia's simulation technologies, we have closed the so-called sim-to-real gap. This paves the way to make physical AI a reality in industry worldwide," says Marc Segura, President of ABB Robotics. "For more than 50 years, ABB Robotics has been driving intelligent industrial automation. We pioneered the development of the first generation of all-electric industrial robots, introduced digital twin simulations using RobotStudio, and launched a new type of autonomous and versatile mobile robot. By working with Nvidia, we are now making the widespread use of physical AI in industry possible."

The collaboration makes it possible to combine ABB Robotics' RobotStudio programming and simulation suite with the simulation capabilities of Nvidia Omniverse libraries, which can perform physically realistic simulations. This combination closes the long-standing technological "sim-to-real" gap. Developers are able to model robots as digital twins in a simulation and collect synthetic data that can be used to train physical AI models. This technology enables companies, regardless of their size or industry, to use AI-driven robotics for various industrial workflows.

The resulting realistic simulations and foundation models are continuously optimized in 'RobotStudio HyperReality'. The system is continuously improved through feedback from the field. The foundation models can be used to train any number of ABB robots anywhere in the world with the reliability and accuracy required by the industry.

"The industrial sector needs physically accurate simulations to bridge the gap between virtual training and real-world deployment of AI-driven robotics across the board," said Deepu Talla, Vice President of Robotics and Edge AI at Nvidia. "The integration of Nvidia Omniverse libraries into RobotStudio combines our advanced simulation and computing power with ABB Robotics' unique virtual control technology. This accelerates the process by which manufacturers of all sizes bring complex products to market."

Closing the "sim-to-real" gap

Until now, there has been a crucial difference between theory on the computer and practice in the factory: the "sim-to-real" gap. Differences in light or materials made it difficult to plan production processes purely virtually.

ABB Robotics is now closing this gap together with Nvidia. Thanks to high-precision simulations and synthetic data in Nvidia Omniverse, intelligent robots achieve a 99 percent match between model and reality. ABB is the only company to have a so-called virtual controller, an exact copy of the original software that controls the robot. As a result, simulation and real performance match almost perfectly. Together with Absolute Accuracy technology, which minimizes inaccuracies from the original 8 to 15 mm to 0.5 mm, ABB offers a level of precision that was previously unthinkable in the virtual world.

Manufacturers can now design, test and optimize production lines purely virtually. The result is major efficiency gains: set-up and commissioning times are reduced by up to 80 percent, while the elimination of prototypes cuts costs by up to 40 percent. In addition, the time to market for complex products, such as in consumer electronics, is halved.1

ABB Robotics is also looking into integrating Nvidia's Jetson edge computing platform into its OmniCore controller. The aim is to make real-time AI available directly on the robot for the entire portfolio.

Real applications today

'RobotStudio HyperReality' supports industrial customers of all sizes and sectors in a wide range of applications. The first pilot customers are already testing the functions ahead of the official market launch, which will take place in the second half of 2026 for all 60,000 ABB Robotics RobotStudio users.

Foxconn, the world's largest contract manufacturer of electronics, is piloting the first joint use case in the production of consumer electronics. Automating the assembly of small parts in this area is particularly challenging, as different model variants require different production methods. In addition, the delicate metal structure of filigree control elements requires high-precision control during gripping and assembly, which until now has often led to time-consuming debugging and high use of engineering resources. With the help of RobotStudio HyperReality, Foxconn now trains its assembly robots virtually. By using synthetic data, real production processes are perfected in various scenarios before being transferred to the production line with an accuracy of 99 percent. This virtual optimization reduces set-up times and costs by eliminating the need for physical testing. This significantly accelerates the market launch of new consumer electronics.

"In consumer electronics manufacturing, precision is everything. Until now, this level of accuracy and detail was simply not achievable in simulations and digital twins," explains Dr. Zhe Shi, Chief Digital Officer of Foxconn. "We see enormous potential in the collaboration between ABB Robotics and Nvidia. Thanks to advanced AI inference, we can parallelize engineering processes, ramp up production faster and significantly accelerate our overall product development."