Factory automation has always carried a costly contradiction. Manufacturers want robots to handle more variable work, yet proving that a new system will perform reliably often requires physical prototypes, repeated commissioning and long periods of engineering support on the production floor.
ABB Robotics and Nvidia believe higher-fidelity simulation can remove much of that friction. Their RobotStudio HyperReality product, scheduled for release in the second half of 2026, embeds Nvidia Omniverse libraries in ABB’s established RobotStudio environment. It is designed to let engineers build and validate a complete automation cell before installing hardware.
ABB says the approach can reduce deployment costs by up to 40% and shorten time to market by as much as 50%. Those gains depend on whether the digital representation behaves closely enough to the eventual factory system.
A virtual cell mirrors the physical one
Engineers can export a fully parameterised station—including robots, sensors, lighting, parts and kinematics—as a Universal Scene Description file into the Omniverse environment. A virtual controller then runs the same firmware used by the physical machine.
ABB claims this produces a 99% behavioural match between digital and physical systems. That level of consistency allows teams to test sequences, reach, timing and interactions before equipment arrives at a site. It also reduces the risk that an apparently successful simulation will fail when exposed to real controller behaviour.
Vision models can be trained with synthetic images generated inside the environment instead of relying entirely on pictures collected from a working line. Varying lighting, object position and part appearance helps prepare the model for conditions that would be expensive or disruptive to reproduce physically.
When paired with ABB’s Absolute Accuracy technology, the companies say positioning errors can fall from 8–15 millimetres to about 0.5 millimetres. That difference determines whether a robot can move a large object approximately or assemble a delicate component consistently.
Early users target faster product changeovers
Foxconn is testing the software for consumer-device assembly, where products change frequently and reflective metal parts can challenge computer vision. Virtual training could help the manufacturer prepare a cell for a new design without stopping a physical line for prolonged testing.
California automation provider Workr is combining its WorkrCore platform with ABB hardware trained through Omniverse. It plans to demonstrate systems capable of onboarding unfamiliar parts in minutes without specialist programming. If that performance transfers beyond a demonstration, it would address one of automation’s largest economic barriers: the engineering effort required every time a product changes.
ABB is also evaluating Nvidia’s Jetson edge platform for its Omnicore controllers. Running inference close to the robot can reduce latency and keep production moving when cloud connectivity is limited. More importantly, it creates a route for existing robot fleets to use newer perception models without replacing the complete control system.
The source estimates that digital-first simulation can cut setup and commissioning time by up to 80%. Results will vary by factory, but the direction is clear: more engineering work is shifting away from scarce production time and into a reusable virtual environment.
ROI depends on disciplined simulation practices
Simulation does not make physical validation unnecessary. Models of friction, flexible materials, wear and unexpected human behaviour remain imperfect. A manufacturer still needs acceptance testing and safe procedures for conditions the digital twin did not anticipate.
The advantage is that physical testing can concentrate on exceptions rather than basic programming. Teams can discover collisions, visibility problems and cycle-time constraints earlier, when changing a virtual layout costs far less than moving installed equipment.
Achieving that benefit requires good data and new skills. Engineers must know how to create representative synthetic scenes, maintain accurate asset definitions and measure where virtual performance diverges from reality. A beautiful simulation with incorrect physics can create misplaced confidence.
RobotStudio HyperReality reflects physical AI’s transition from research to industrial economics. The question is no longer only whether a robot can learn a task, but how quickly a company can move that task from experiment to dependable production.
By linking controller-accurate simulation with synthetic training, ABB and Nvidia are attempting to make deployment repeatable. If manufacturers can reuse virtual cells, test more scenarios and commission equipment with fewer surprises, the return comes from avoided disruption as much as faster robot motion. That is the practical threshold physical AI must cross to become ordinary factory infrastructure.
