AI, XR, digital twins set to transform robotics

The arrival of embodied AI

The advent of embodied AI marks a step towards making AI available across applications. As defined by Nvidia, embodied AI represents “the integration of artificial intelligence into physical systems, enabling them to interact with the physical world”.

Nvidia, which is at the centre of AI developments with its enabling chips, notes: “The fusion of machine learning, sensors and computer vision lets … systems perceive, reason and act in real-world environments.”

Embodied AI goes beyond robotics in the strict sense and applies to smart systems and infrastructures more generally. It extends the capabilities of AI to physical systems – such as buildings, robots and autonomous vehicles like cars, trucks and robotaxis – and by integrating machine learning and computer vision, these systems can unlock the potential of generative AI applications in physical industries.

AI models can leverage data that existing robots collect will operating. These models then inform robotic applications by bridging the gap between simulations and real-world applications. In this context, digital twins will play an important role. These will provide synthetic data that can supplement data collected in the field. This type of data is artificially created data “designed to mimic real-world data,” says IT giant IBM.

Splicing virtual and physical applications

Industrial-machinery manufacturer Siemens is looking at the benefits digital twins have to offer for integrators and users of robots and industrial equipment.

Brian McMinn, machine tool business segment manager at Siemens, explains how such virtual environments support computer-numerical-control (CNC) machines in an increasingly digitalised world: “We can dry run in a virtual world before they even start building the machine.”

Industrial-robots producer KUKA similarly leverages digital twins to support its product offer. For example, stove manufacturer HASE Kaminofenbau uses digital twins and KUKA welding robots in its operations.

Florian Fischer, head of production development at HASE, outlines the advantages: “We wanted to build an ultra-modern, flexible robotic system that will also be able to process future models that don’t even exist yet, without placing constraints on our designers or bringing production to a standstill.”

Toppan offers its own digital-twin solution TransBots to address the changes digitalisation will effect across all spheres of life. The global printing and packaging company applies a very wide view on the future use of digital twins, stating: “As our society is going to be more and more digitised in the future, we will need to implement a digital twin system where information is shared between humans, robots and services in a virtual space, thereby allowing work to be performed efficiently.”

Station Ai in Nagoya, Japan, is using a digital twin solution to establish a “robot-friendly environment” within its open-innovation-focused facility with a participating network of “more than 1,000 startups, partner companies, VCs and other support organisations, and universities”, according to the firm.

Training robots in digital twins

Digital twins can not only explore layouts and workflows that accommodate for robotic systems, but can provide training grounds for robotic systems to accelerate their use across application areas and lower cost associated with robotic applications. Digital twins can overcome hurdles that currently prevent the use of AI in some applications.

In manufacturing, companies can collect and analyse data from the factory floor to train AI-enabled robotic systems. But many manufacturers lack such data-collection abilities. Similarly, in fairly unstructured operations such as in mining, but also many manufacturing environments (including construction), the complexity of human operations can quickly outstrip the benefits of integrating AI-powered machinery or “cobots”, collaborative robots.

In most manufacturing facilities, digital twins can create synthetic data to train AI-enhanced robotic equipment. In more complex, rapidly changing surroundings, digital twins offer a pathway to facilitate the use of AI systems in the future, particularly if these twins can feed on real-time sensor data that reflect ongoing changes.

Pre-training of AI models with actual data from exiting robotic systems can create reliable and robust models to train models that improve on operations. Sensor-based digital twins that mirror real-world behaviour of systems and workforce can offer a viable alternative. But digital twins can go further and provide training grounds in environments with low data availability. Simulations within virtual representations provide an option to transfer real-world dynamics into virtual environments to then train AI, robotics, and equipment for real-world applications.

“Synthetic data, generated from digital twin simulations, can be used alongside real-world data to train multimodal physical AI models,” says Nvidia. “Synthetic data generation is the creation of text, 2D or 3D images, and videos in the visual and non-visual spectrum using computer simulations, generative AI models, or a combination of the two.”

For example, roboticists can leverage now digital twins to quickly create scalable environments that support the training and optimisation of AI models that can then find use for training robotic navigation or vision systems, to name just a few use cases. Such simulations not only accelerate training of robotic and automated systems but also allow developers to consider scenarios that can be difficult or prohibitively expensive to create in real-world environments.

Moreover, such simulated environments can inform AI and robotic systems about the effects of potential disruptions that are otherwise impossible to create. For example, the effects of events such as the Covid pandemic – which turned demand patterns upside down, put tremendous stress on supply chains, and changed the flow of people and goods globally and regionally – can only be simulated in virtual environment to test AI models.

Similarly, strategists and modellers can now consider the changes natural disasters, conflict outbreaks, or major competitive actions can have on the commercial environment at levels that cannot be reasonably replicated in any other ways. Simulations in digital twins can guide thinking at global and geopolitical levels.

AI supports XR, XR accommodates AI

On a related, forward-looking sidenote, consumers will increasingly adopt home robotics in their households. Meta Platforms CEO Mark Zuckerberg also sees AI as an enabler for virtual environments. After Meta focused efforts on the creation of metaverse computing environments a few years ago, AI quickly moved to the forefront of media and investors’ attention, supplanting interest in metaverse applications.

Now, Zuckerberg considers AI an important part of these efforts as he outlined on 17 September 2025 at the Meta Connect 2025’s opening keynote, in which he presented Meta's vision for AI and the metaverse.

The combination of AI and XR elements will find use to create content easily and to make it look more authentic to the real world. Meta Horizon Studio is Meta’s suite to “build and iterate [content] fast with generative AI and ship to a global audience on mobile and VR” in the metaverse.

Zuckerberg stated: “Soon, Meta Horizon Studio is going to include an agentic AI assistant that will stitch together…different tools and further speed up the creation process using just simple text prompts … Now you are going to be able to easily create infinite connected spaces that look way better with realistic physics and interaction.”

Familiarity with such environments will enable consumers in the future to create their very own digital twins of homes and backyards, for instance. Such twins then provide a platform to operate home robots.

Four technologies enable exponential progress

At the beginning of the internet economy in the 1990s, many companies started by initially posting catalogue-like information on websites. Over time, ordering and purchasing processes followed. By now, internal, external and partnership processes use internet connectivity as the glue that connects commercial activities and enables more and more use cases – from desktop uses to mobile applications to automatic operations.

Similarly, sensor and synthetic data, AI, robotics, digital twins and XR will diffuse over time across application areas. In time, digital twins and XR will become environments that users won’t even distinguish from real-world environments, similar to the way that internet applications now are woven into virtually every aspect of our lives.

Advanced robotics will become increasingly autonomous in industrial applications, and even consumers will learn to consider them appliances the way we see internet-connected refrigerators and cooking appliances – robotic vacuum cleaners and lawn mowers already have carved out valuable markets.

Over time, the combination of all these technologies will meld into powerful applications, similarly to the way internet applications, mobile devices, cars and appliances now combine seamlessly.

Martin Schwirn is the author of Small data, big disruptions: How to spot signals of change and manage uncertainty (ISBN 9781632651921). He is also senior adviser for strategic foresight at Business Finland, helping startups and incumbents to find their position in tomorrow’s marketplace.