ABB Robotics and NVIDIA have published a joint white paper describing how physical artificial intelligence (AI) could be applied in precision manufacturing through a digital-first engineering approach that combines simulation, synthetic data and real-world operational feedback. The paper argues that traditional industrial automation is often too rigid to adapt quickly to changing manufacturing environments and proposes a methodology aimed at accelerating the deployment of AI-enabled robotics.
The companies describe physical AI as a combination of robotics, AI models, simulation and accelerated computing that allows robots to perceive, adapt and learn from their environments. A central element of the proposed approach is the use of digital twins to validate robotic vision systems and AI models before physical deployment, enabling manufacturers to identify potential issues during the design phase rather than after installation. Operational data collected from deployed robots would then be used to refine digital models through a continuous feedback loop.
The white paper identifies robotic vision as a key entry point for broader physical AI adoption in manufacturing. It outlines an engineering process that incorporates task-specific synthetic data, AI validation and robotic verification to create reusable and traceable engineering assets intended to reduce deployment risk and improve scalability.
ABB also describes its Physical AI Toolchain, which is designed to support training robots using simulated, synthetic and real-world data rather than relying solely on conventional programming. According to the company, the toolchain is intended to allow manufacturers to integrate different AI models and datasets while maintaining industrial-grade performance requirements.
The publication builds on the partnership announced by ABB Robotics and NVIDIA earlier in 2026. One of the first products resulting from that collaboration is RobotStudio HyperReality, which integrates ABB’s RobotStudio simulation software with NVIDIA Omniverse technologies to reduce differences between simulated robot training and real-world operation. The companies have said the platform is expected to become available to RobotStudio users during the second half of 2026 following customer trials.
The white paper was developed with contributions from AsiaInfo, Deloitte and SKAI Intelligence.
