The 21st-century industrial transformation is not solely measured by robots, automated factories, or artificial intelligence algorithms. Increasingly, the key lies in a concept that until recently seemed like science fiction: digital twins. Virtual replicas, hyper-realistic and dynamic, of environments, processes, and physical products that enable experimentation, testing, and optimization without any risk.
This week, during the SIGGRAPH conference, NVIDIA unveiled a set of innovations aimed at taking this technology to the next level. Under the umbrella of Omniverse and the open standard OpenUSD, the company announced new developments in simulation, rendering libraries, foundational world models (World Foundation Models, WFMs), and certifications that seek to promote the global adoption of digital twins across key industries.
Even though the term was coined over two decades ago, digital twins have quickly evolved from a promise to a common tool in sectors such as automotive, energy, logistics, healthcare, and advanced manufacturing.
The idea is simple but powerful: create an identical virtual representation of a factory, supply chain, or even a city. This copy allows for testing scenarios that would be too costly, slow, or dangerous in real life. What happens if the design of an assembly line is changed? How will an autonomous drone behave in a real urban environment? What would be the impact of modifying a logistics protocol at an international port?
All of this can be simulated in real time, drastically reducing costs and speeding up decision-making. Additionally, these simulations serve as the perfect training ground for physical artificial intelligence, from industrial robots to autonomous vehicles.
One of the main challenges of digital twins has always been technical complexity. Integrating 3D data from diverse sources—engineering blueprints, IoT sensors, CAD models, automation systems—into a coherent and realistic environment is a monumental task.
This is where OpenUSD (Universal Scene Description) comes into play, an open standard originally developed by Pixar and now supported by the Alliance for OpenUSD (AOUSD), of which NVIDIA is an active member.
OpenUSD acts as a “common language” for describing complex 3D worlds. It enables multiple applications and teams to collaborate smoothly—an essential feature in industrial projects involving manufacturers, software providers, engineers, and operators across different countries.
The alliance has recently added participants such as Accenture, Esri, HCLTech, PTC, Renault, and Tech Soft 3D, confirming the ecosystem’s strengthening.
During SIGGRAPH, NVIDIA showcased a range of tools and advancements focused on accelerating both the development of digital twins and their application in physical AI:
Omniverse SDKs and libraries: direct connection between MuJoCo and OpenUSD, opening opportunities for over 250,000 robotics developers to simulate machinery in realistic environments.
Omniverse NuRec: neural reconstruction libraries and AI models capable of generating realistic 3D environments from sensor data, utilizing advanced Gaussian splatting techniques powered by RTX.
Isaac Sim 5.0 and Isaac Lab 2.2: simulation and learning frameworks for robots, now with neural rendering and expanded sensor and robot schemas in OpenUSD, available as open-source projects on GitHub.
Cosmos WFMs: foundational world models that improve synthetic data generation and physical AI reasoning, including Cosmos Transfer-2 and Cosmos Reason.
Advances in rendering and materials: applied research to automate the creation of photorealistic materials and boost the scalability of digital twins.
The announcement also includes official certification for OpenUSD development and free training programs on digital twins, reinforcing the educational and professional development aspects.
Beyond theory, leading companies are already deploying these technologies:
Siemens has integrated Omniverse and OpenUSD into its Teamcenter Digital Reality viewer, allowing engineers worldwide to collaborate on photorealistic digital twins, speeding up design reviews and reducing physical prototyping.
Sight Machine combines live production data, generative AI, and digital twins to provide plant teams with instant visibility and automated recommendations.
Rockwell Automation uses the Emulate3D Factory Test platform to build scaled virtual factories, simulating automation systems and validating processes prior to actual deployment.
EDAG employs industrial digital twins to optimize production designs, train workers, and perform data-driven quality control.
Amazon Devices & Services trains robotic arms in virtual environments to recognize and assemble devices, minimizing risks and speeding up new product launches.
Vention integrates Omniverse, Isaac Sim, and Jetson hardware to offer plug-and-play automation solutions with digital twins, simplifying smart factory deployment.
A less visible but equally critical aspect is cybersecurity. The widespread deployment of digital twins involves handling vast amounts of sensitive industrial data, raising risks such as industrial espionage, digital sabotage, or manipulation of simulated environments. Experts emphasize integrating cybersecurity from the design phase—using encryption, access controls, network segmentation, and continuous monitoring—to prevent these virtual environments from becoming vulnerabilities.
The launch of certifications and free courses by NVIDIA addresses the pressing need to train professionals capable of creating, managing, and safeguarding digital twins. The skills gap in AI, simulation, and cybersecurity is becoming more evident; without prepared teams, technological advances risk stagnation in labs rather than real-world application.
The combination of digital twins and physical AI promises a revolution comparable to industrial automation in the 20th century. Robots trained entirely in virtual worlds will operate safely in real environments; cities will optimize traffic flows through simulations before implementing changes; factories will adjust production lines in real time using predictive models.
NVIDIA’s message is clear: we are entering an era of AI applied to the physical world, with digital twins as the key tool to make the leap.
What NVIDIA showcased at SIGGRAPH are not just technical updates but the foundation of a new way of rethinking industry, logistics, and technological innovation. By adopting OpenUSD as an open standard, the company aims to democratize access, avoid proprietary lock-ins, and foster global collaboration.
In an increasingly digital world where the boundary between physical and virtual blurs, digital twins could become the catalyst for the next industrial revolution driven by artificial intelligence.
Frequently Asked Questions (FAQs)
What is a digital twin, and what is it used for?
A digital twin is a virtual replica of a physical object, process, or environment. It’s used to simulate, test, and optimize without risking real resources.What role does OpenUSD play in this revolution?
OpenUSD is an open standard that describes and connects complex 3D worlds. It facilitates collaboration across platforms and companies, serving as the technical backbone for scaling digital twins.Which companies are already using digital twins with NVIDIA Omniverse?
Siemens, Amazon, Rockwell Automation, EDAG, Sight Machine, and Vention are among those applying Omniverse and OpenUSD in design, manufacturing, and automation processes.What security risks are associated with digital twins?
The main concern involves protecting industrial and strategic data. Compromise or manipulation of a digital twin could affect both simulations and real operations, making cybersecurity essential.