The final vision of an autonomous factory
Physical AI has the potential to transform automotive manufacturing depending on globally recognised data standards and trust in the tools being developed with them, according to speakers from Hyundai, Lucid, Schaeffler and Accenture at this year’s Nvidia GTC conference
Digital platforms are enabling OEMs to visualise production processes end-to-end before executing them on the shopfloor. The challenge now is to get the workforce of new talent up to scale and coherent with the latest technology and the speed of its development.
The deployment of more intelligent automation is being coordinated by [AI] agents and that is what will ultimately be the final vision of an autonomous factory
At this year’s Nvidia GTC global AI conference, speakers from Hyundai, Lucid, Schaeffler and Accenture looked at how the automotive industry is being transformed by AI and how a digital twin of the supply chain is critical to optimising manufacturing decisions. The journey from generative AI for design to agentic AI for orchestrating production in complex environments is now entering the phase of physical AI, driving fundamental change in the real world by simulating manufacturing processes and then deploying autonomous robotics to carry them out.
“The deployment of more intelligent automation is being coordinated by [AI] agents and that is what will ultimately be the final vision of an autonomous factory,” said Alpesh Patel, senior vice-president of the Software-Defined Factory Transformation Center at Hyundai Motor Group.
Software-defined factory
Hyundai Motor Group is currently building an AI factory with Nvidia in Korea to accelerate model training, validation and deployment for in-vehicle AI, autonomous driving, smart factories and robotics. Nvidia provides hardware and software for AI and will Hyundai will be deploying Nvidia’s Blackwell AI infrastructure across manufacturing and robotics, as well as autonomous driving. The two companies are working with the Korean government to develop the country’s AI infrastructure. That includes the establishment of an AI application centre and an AI technology centre, while developing local talent. Hyundai is also exploring the use of Nvidia Omniverse and Cosmos platforms on Nvidia RTX PRO Servers to develop car factory, digital twins and robots.
The Software-Defined Factory is Hyundai’s initiative to make manufacturing automation more intelligent and take on physical tasks that were previously inaccessible, including those laborious and repetitive tasks that workers no longer want to do. Physical AI plays a major role in this initiative and what is, according to Patel, is the integration of the supply chain and the creation of a digital twin of that supply chain from tier-n production all through to aftersales. With everything connected, Hyundai will use relevant AI agents to optimise decisions at a macro level across operations and at a micro level within each respective factory.
The challenge with bringing on each factory in Hyundai’s global network is the different ages of those factories, making middleware platforms crucial to the success of the enterprise. The hardware and software systems that directly monitor, control and manage physical devices, processes and production lines need to be comprehensive. Patel said that operating a portfolio of factories of very different ages means making sure the data provided can be understood by the AI agents deployed whether the equipment in those factories is 25+ years old or not.
“I think that those foundational technologies on the operational technology layer are extremely important to what we are trying to put as an overall supply chain,” he said.
Roberto Henkel, senior vice-president of digitalisation and operations IT at tier one components supplier Schaeffler, also talked about the need for a standardised platform to optimise daily operations across factories old and new. Henkel said roughly 20 out of its 100 factories were less than ten years old and there is a huge challenge to establish the standardised platform necessary to gather data from the older facilities so Schaeffler can execute digital tools at scale.
For older plants it is about enriching data with context and building a virtual foundation on which to make decisions about how to make manufacturing more efficient.
Schaeffler is working on a new factory where it can use Nvidia Omniverse, a multi-GPU, real-time simulation and collaboration platform, for new product launches. However, he pointed to project required to bring the 80 older factories onto the platform to access data and in doing so provide new capabilities to existing plants by enriching data with context and making decisions based on that information. Henkel said Schaeffler needs to optimise existing brownfield plants and scale the capability of its manufacturing network, showing in pilot factories what is possible and then bringing in legacy facilities.
Digital workbench
Tracey Countryman, global supply chain and engineering lead at technology consultancy Accenture, pointed to the need for companies to consider the tools they are using to put brownfield equipment in the cloud.
“What about the workbench? What are the tools my people are going to use? What are the security and frameworks that are going to sit underneath to govern AI and the guardrails that need to exist? she asked.
Countryman said that most sites were never intended to be connected other than in their closed loop, level 2, manufacturing execution system. That brings with it a whole new risk profile for cyber security and the need for heightened IT governance.
According to Countryman physical AI is going to be exponentially more impactful because it enables companies to drive fundamental change in the physical world by simulating operations first. She pointed to one Accenture client that has been using digital twins in production for eight years and has connected 20,000 machines across 20 facilities, across nine countries. It now has the ability to monitor, control and now simulate by adding in agentic in the latest releases. “It is fundamentally changing how they think about their value chain and how they drive optimisation, and ultimately agility and growth,” said Countryman. “That is not something we would have seen in recent years but now with the technology stack that is going to accelerate even more. That is the next wave of growth around digital twin.”
“You can go anywhere from humanoid robotics to AI vision, to collecting operational data and predicting defects or downtime before it happens. This lends itself now to be all put in one ecosystem, in [Nvidia’s] Omniverse, Isaac Sim or Metropolis so we can really be able to integrate all of this work in one platform
Ecosystem for EVs
As a relatively new electric vehicle maker, Lucid benefits from state-of-the-art manufacturing, with one US car plant running in Casa Grande, Arizona and another in Saudi Arabia in King Abdullah Economic City, which is now working up to full production from kit assembly. It also has a powertrain plant in Newark, California.
The Saudi facility is equipped with the latest controls, equipment and technology, according to Jason Ryska, vice-president of manufacturing engineering, Lucid Motors. Its powertrain operations are automated with the latest technology and with “a ton of data transfer from operation to operation”, said Ryska, but the carmaker now needs to bring everything together on one platform.
“You can go anywhere from humanoid robotics to AI vision, to collecting operational data and predicting defects or downtime before it happens,” said Ryska. “This lends itself now to be all put in one ecosystem, in [Nvidia’s] Omniverse, Isaac Sim or Metropolis so we can really be able to integrate all of this work in one platform.”
Ryska stressed the need for a common data model format. “Think about that across the enterprise and what you are trying to accomplish and set that up first and the AI, analytics and the modelling become much easier and much quicker,” he said.
Universal data
Hyundai’s Patel said the global standardisation of the data generated from manufacturing and supply chain for AI applications was integral to Hyundai’s Single Data Model. Beginning with manufacturing but then extending to the downstream supply chain, Hyundai’s goal is to establish a universal data standard that is understood globally by everybody in Hyundai’s operations and those of its suppliers. With that it will be able scale the AI applications based on the data. Without it, Patel said that proof of concepts would only be irregular and piecemeal, with a return on investment hard to meet down the road.
Those data models need to be built for each individual macro process and that is a very important piece of work,” he said. “It is not only at the level of the platform whereby we operate the agents but it starts at the machine, the operational technology level, and needs to move all the way.”
Making that happen requires a huge culture shift in various parts of the company, according to Patel, and he called it a double transformation.
“We have to transform the way in which we are deploying the solutions in the factories from purely looking at a manufacturing/engineering standpoint to a process and data related standpoint,” he said.
Production operatives also need to be made aware of the data and what it means. That explanation is critical in building intuition and trust in the physical AI system built on it.
Trusting the tools
What is as important as making data coherent across the supply chain and across manufacturing plant networks globally is the need for a tech savvy workforce that trust the data and the tools they are increasingly expected to work with. A new generation of college graduates that are technically literate are pushing the automotive organisations they are jointing to innovate with digital technology in a way that only the lonely scientist in the room would have done previously. However, those companies need to rethink how they make AI part of their operating fabric, according to Tracey Countryman.
Companies need to fundamentally rethink the work process from beginning to end, including who does it and how it gets done. That includes simulating the planning and scheduling of production work at the plants. Everyone working in automotive production will require some level of AI and data fluency to leverage the models they are using but they also need to understand the rationale behind using the technology.
“If I think about one of the hardest things in operations and engineering it is understanding the rationale and the logic…[.] You have to explain that and engineers need to understand what the decision is and trust that the model is doing what it needs to do,” said Countryman.
Schaeffler’s Henkel also pointed to the importance of trust. “The first reaction if you confront someone with an issue in the production line that was made obvious by data is that the data is wrong,” he observed. “We need to overcome this.”
To do so means making the way the data is delivered, standardised and presented reasonable and valid for the user and in that way build trust in the decisions made based on that data.
“This is why we have started a huge initiative of bringing data quality and making it visible to top management. We have an activity that everybody feels responsible for the data they are generating and providing through all the technology we have for broader usage.”
A more trusting mindset goes hand in hand with reskilling people because until AI is able to have jurisdiction on objects and processes, people are going to be ones that actually have clarity to make decisions on the job and set the guardrails for safety. AI agents and simulation tools will model processes and provide intelligence on completing routine tasks but the really complex tasks will still need intervention and control by the upskilled worker. “That is why it is not about humans in the loop with AI, it is actually humans in the lead,” said Countryman.
Ryska said that fluency in data analytics, AI tools and methodology is increasing and more people are coming into vehicle manufacturing that are able to deploy tools in their own work environment.
“We have the responsibility to new college graduates to set up the environment where they can use that skillset and do things that are even greater and beyond what we’re thinking about now.”
