ABB Robotics and NVIDIA bring Omniverse into RobotStudio for industrial physical AI
Original: ABB Robotics Taps NVIDIA Omniverse to Deliver Industrial-Grade Physical AI at Scale View original →
ABB Robotics and NVIDIA said on March 9, 2026 that they are integrating NVIDIA Omniverse libraries into ABB's RobotStudio suite to bring physically accurate simulation and synthetic data generation into mainstream industrial robotics workflows. The resulting product, RobotStudio HyperReality, is scheduled for the second half of 2026.
The companies frame the announcement as an attempt to close the long-standing sim-to-real gap. ABB says the combined system can reach 99% correlation between simulation and real-world robot behavior because the virtual controller runs the same firmware as the physical robot. The product exports robot stations as USD into Omniverse, where engineers can test robots, sensors, lighting, kinematics, and parts before any physical deployment.
- ABB says engineering time can fall substantially.
- Deployment costs are projected to drop by up to 40%.
- Time to market is projected to improve by as much as 50%.
- Setup and commissioning time can fall by up to 80% in virtualized production-line validation.
ABB says more than 60,000 robotics engineers use RobotStudio, which makes this less of a lab demo and more of a platform-level move. Early pilots include Foxconn in consumer electronics assembly and Workr in small and medium-size manufacturing. NVIDIA also says ABB is exploring Jetson integration into its Omnicore controller for real-time inference across the robot portfolio.
The importance of the deal is that industrial AI deployment usually fails on data quality, calibration, and validation. If the 99% sim-to-real claim holds in production, manufacturers get a faster way to train vision models with synthetic data, validate cells before installation, and ship robotic automation without the usual amount of physical trial and error.
Related Articles
NVIDIA released open-source physical AI agent skills across Omniverse, Cosmos, Isaac, Metropolis, Alpamayo, and Jetson. The company points to manufacturing gains including 67% faster training and deployment at Pegatron and a 17% detection-rate improvement at Delta Electronics.
NVIDIA released Cosmos 3 as an open physical AI omnimodel with Super and Nano variants. Its technical post points to six synthetic datasets, Hugging Face checkpoints, and GitHub recipes for domain adaptation.
NVIDIA’s open humanoid reference design combines Unitree H2 Plus hardware, Sharpa five-finger hands, and Jetson AGX Thor T5000 compute. The 75-DoF system is aimed at making humanoid research more comparable across labs.