Though the details can obviously be complex, the concept of using AI with digital information to solve problems is straightforward. But what about physical information? More specifically from a business standpoint, what about using AI to solve challenges in complex physical environments that are constantly in flux, like warehouses? That’s the challenge that Nvidia, Accenture and the big supply chain automation company KION have partnered to address.
Last week at CES I had a chance to sit with Accenture CEO Julie Sweet, KION CEO Rob Smith and Nvidia CEO Jensen Huang to talk about their partnership during a special session for four press outlets. (I was the only analyst in attendance.) I believe that in the long run technology like this will fundamentally change the way supply chains work. There have been many fits and starts with “industrial IoT” and “Industry 4.0,” however, and we don’t know yet how long that will take.
Using Physical AI To Bring More Order To The Complexity Of Warehouses
Warehouse logistics within big supply chains can be insanely complex. Workers and managers have to factor in an endless array of constantly shifting variables, from consumer demand to inventory on hand to weather conditions. This trillion-dollar market touches every sector that handles physical goods — healthcare, electronics, food, CPG, you name it — each with its own special considerations that must be factored in. It’s all so complex that accurately predicting operational performance for warehouses and distribution centers can be next to impossible. As Smith, who has decades of experience in this space, put it, “Everything’s changing all the time in a warehouse, and systems aren’t smart enough to not only figure out what’s the next move, but figure out what could be potential other next moves and what’s the best next move.”
Now, Accenture, KION and Nvidia are promising a way out of this maze. Physical information from a warehouse can be digitalized by KION software into a highly accurate digital twin that lives on Nvidia’s Omniverse platform. Nvidia’s AI technology then enables rapid simulation of scenarios under different conditions to optimize warehouse operations. Accenture applies its expertise to help define and manage KPIs.
While digital twins have been used for years, they’ve never had this much horsepower across multiple layers of cutting-edge technology — and especially not this much advanced AI. The business rationale is clear. As Huang said, “Every industry that becomes digitalized moves faster, [and] everything you can software-define becomes more capable. . . . When you’re digitalized, you can build consistently with greater capability, but when you become software-defined on top of that, you get to revolutionize your business.”
Huang pointed out that these advantages have been commonplace in the IT industry for decades, but have never been enjoyed broadly in the industrial sector. This makes sense because it’s easy to take digital data or a digital product like a computer or a microchip and then use digital tools to manage it. But, Huang said, “The world’s physical plants, the physical world, has never been digitalized, truly — not until now.”
The Physical AI Model And Software Under The Hood
To dig into the specifics, KION is adopting Mega, an Nvidia Omniverse blueprint for large-scale industrial digital twins. Underlying this is Nvidia’s Cosmos physical AI model. “The fundamental idea of Cosmos,” Huang said, “is a model that understands the physical world like ChatGPT understands information and language.” Huang compared Cosmos to Meta’s Llama and OpenAI’s GPT-4, noting that Cosmos was trained on 9 trillion parameters — a process that required six months and tens of millions of dollars of investment. He said that Nvidia is making it an open model “like Meta opened Llama.”
As with ChatGPT, Cosmos allows a user to generate a bunch of alternative outputs. But whereas ChaptGPT could write you many different versions of a fairy tale, Cosmos can generate many different 3-D video simulations of a specific warehouse under alternative scenarios. These can account for different layouts, numbers of employees, numbers of robots, and so on to allow facility operators to understand which scenario is best for throughput, labor cost, safety measures, error rates, or whatever other KPI is desired. To make sure the simulated versions don’t contain any hallucinations, the system is grounded in the real-world context of the facility, supplied via Omniverse. This data comes from still and video images, CAD models, lidar scans, sensors on robots and other sources to anchor the scenario in the exact physical details. Huang compares this to using RAG to prevent hallucinations in ChatGPT or the other non-physical AI environments we’re more familiar with. It took me a lot of time and research to distinguish between Cosmos and Omniverse, and I like to think that Omniverse makes Cosmos results more accurate, like RAG does for many other non-physical enterprise AI applications.
Because facility operators cannot afford downtime, all of this happens on the fly. KION’s warehouse management software assigns a task — say, moving a load from one location to another — and the industrial AI “brains” within the digital twin work out the implications, planning and (virtually) carrying out next steps while Mega tracks what happens through continuous feedback loops. All of this can be simulated as much or as little as needed within the digital twin to optimize for specific outcomes, then implemented in the real world. During the small group session, Smith pointed out the importance of this for handling different real-time conditions inside a distribution center, using the contrasting examples of Black Friday and a slow summer day.
Real-World Impacts And The Labor Market
As for the impacts of this technology, Sweet believes that it could ultimately cut the time it takes to plan a new warehouse in half. For ongoing operations, she projects similar 50% reductions in manual labor and operating costs. She (echoed by the other two CEOs) expressed that this could take pressure off the retail industry, consumer goods makers and other sectors that have been battered by inflation in recent years. She added that this initiative “is very much about resilience and agility” for the companies that will use it.
The CEOs all agreed that it can also help address ongoing labor shortages. “The fact of the matter is,” Huang said, “we’re tens [of millions] or 100 million workers short around the world. We are deprived of revenues because of worker shortage, not the other way around. There’s several trillion dollars’ worth of lost revenues because there aren’t enough workers, and so we need to augment the workers that we have.”
Smith gave more color on this for warehousing in particular. He said, “It’s very difficult for all of our customers — worldwide, any region, every segment, every vertical — [they] can’t find manual labor to come in and work in a distribution center.” In that context, “We’re automating to make every job better, and that gives people opportunity.” Rather than trying to attract workers, especially young people, to manual entry-level jobs, automation takes a lot of those jobs out of the picture while introducing higher-skilled jobs to make sure the systems are running right — jobs that Smith says are much more exciting and interesting.
Are Companies Ready To Adopt This Much Automation?
Last week’s CES was my 20th one to attend, and I can recall bullish announcements from ten years ago about reinventing warehousing, smart transportation and related tech. And I do believe that the technology available to us today — not just the AI but also other aspects such as edge computing — is much better than it was then. But during the session I asked the three CEOs what makes this moment different in terms of uptake for their customer companies. In other words, what other gears besides the existence of better technology need to click into place?
Smith answered by pointing out the great advances in recent years for autonomous mobile robotics. Simply put, there are many more autonomous robots handling tasks in today’s warehouses and factories, which takes us beyond the classic fixed robots that have, for example, been used in automotive assembly for decades. He emphasized how autonomous mobile robots give operators much more flexibility to scale over time — now augmented with real-world scenario planning and advanced orchestration that simply weren’t possible before.
That’s the practical side as it applies on the shop floor. Sweet added to it by speaking to organizational readiness. She said that in the years leading up to 2022, she and her colleagues were telling every client company that it needed to reinvent itself with AI, but that only 20% of client CEOs were aligned with that view. Fast-forward to today, after years of generative AI being so widely touted and used, and “It’s exactly flipped . . . at least 80% of CEOs have embraced AI.” She described it as “a fundamentally different condition than we’ve seen for the last decade,” adding, “We are not out there convincing people.”
Huang also pointed out that the technology itself can help companies get past a chicken-and-egg problem: “In order to automate, you have to make investments, and making that investment is hard to activate unless you can see the returns — but you can’t see the returns until you make the investment.” But a digital twin “allows us to lower the bar” for potential customers to understand those returns. “And so instead of having to build out their factories, automate their factories, before they see the benefits of it, they can simulate their factory and see the benefits of it.”
The Future Is Bright, But The Timeline Is Uncertain
As with any potentially revolutionary technology, I’ll believe the most optimistic projections only after I see some real-world results. We’re not sure when that will be, because so far the three partners haven’t committed to a specific timetable for rollout. But when Smith talks about this kind of automation becoming so prevalent and so advanced that all the robotic nodes in the supply chain talk to each other, I’m prone to believe him. Ditto when he says, “Ultimately, I think everything that has a physical instance is going to have a digital instance as well.” It’s just a question of how soon “ultimately” will get here.
The opportunity is definitely enormous. Smith believes that less than 20% of the world’s warehouses have significant amounts of automation in them today. And if anyone would know, he would, because KION has helped a slew of companies across many different industries — from Amazon on down — with warehouse automation.
What I heard from all three CEOs fits with what we’ve learned about (digital) AI in recent years, and it fits with the robotics advancements that Huang discussed in depth during his CES keynote. So I do believe this kind of automation can help supply chain operators make better decisions informed by real-world conditions, which could raise performance standards and improve efficiency and productivity across highly autonomous — and potentially safer — supply chains. We just don’t know how soon that future will arrive. I am a tech optimist, though, and do believe that the time is now.
