In June 2026, Symbotic Inc. named Dr. James Kuffner as its chief technology officer, succeeding George Dramalis. Kuffner has more than 30 years of experience in engineering research and executive management, and he joined Symbotic as the company has grown to be a leader in warehouse automation.
Founded as CasePick Systems in 2007, Symbotic has combined technologies including automated storage and retrieval systems (ASRS), autonomous shuttles, packaging and palletizing equipment, AI, and automated lift trucks. The Wilmington, Mass.-based company serves major retail, wholesale, and food and beverage companies.
In the past month alone, Symbotic acquired sortation system provider cellumation and expanded its commercial agreement with Walmart. Interact Analysis has named the company “America’s largest warehouse automation vendor.”
After studying at Stanford University and teaching at Carnegie Mellon University, Kuffner worked at Google, Joby Aviation, and Toyota. At Toyota, he was in leadership roles at Woven, Toyota Research Institute, and Toyota Motor Corp.
Automated Warehouse recently spoke with Kuffner about his experience and his outlook for Symbotic.
Symbotic continues to grow to meet demand
Symbotic has moved more than 2 billion cases. How have you seen the company growing?
Kuffner: It has been a really wonderful experience. We’re regularly handling over 10 million cases a day now across networks. I keep telling the team that it’s the most exciting time to be working in robotics and physical intelligence.
We’re on an aggressive growth trajectory. We’re not only trying to improve the technology, but also to expand the customer base with lots of different new verticals, entering medical supply, beverage supply, and other things.
We are seeing a lot of demand for this technology, even outside the U.S. and in places where labor is relatively cheap, because of the precision and accuracy that’s possible. And to be honest, these are these are really tough jobs: lifting 60-lb. boxes every day, moving millions of cases of goods so that the supply chain runs efficiently. You get the precision that is really not possible without this automation technology.
You mentioned new markets for expansion. How do you look at the needs in these new verticals?
Kuffner: Moving cases has been Symbotic’s bread and butter since its founding. But sometimes, you don’t want a whole case, and so our break-pack system handles eaches and totes, and our micro-fulfillment solution does native tote handling.
In refrigerated and frozen goods, you have other requirements. Our system has always kept track of expiration dates and how long cases have been in storage, but you have to manage that more carefully and with good traceability for cold chain requirements.
There’s also a lot of complexity in coordinating the whole fleet. It’s no longer just getting robot to work. In some of our larger sites, we have over 1,000 robots, and there are sometimes 1,500 to 1,600 robots moving cases. We also have sites with dozens of robot arms palletizing and depalletizing.
And now we’re really excited about our acquisition of Fox Robotics, which allows us to automate even more of the dock. We think this is another opportunity to make warehouses more efficient. You can turn around trailers faster, allow floor-to-ceiling loading of trucks, and get a 50% more dense pallet. You could actually send the same amount of freight with fewer trucks and save fuel.
Warehouse automation benefits from AI advances, says Kuffner
Despite economic headwinds, robotics adoption continues to grow, according to the latest reports from the International Federation of Robotics (IFR) and the Association for Advancing Automation (A3). How do you see the state of robotics?
Kuffner: It’s a great period of growth and inflection. Part of it, I think, has to do with perception working now much more reliably than it used to due to the advances in algorithmics for machine learning and, of course, our compute infrastructure, our ability to calculate and process vast amounts of data to train large neural networks that can now run in real time and provide real-time perception. This is the unlock for robots operating reliably in semi-structured and unstructured environments.
Previously, all the robots in car painting and welding required perfect models and repeatable tasks. In the case of Symbotic, we have cases that have torn packaging, tape hanging off, or spilled contents. There’s a long tail of edge cases to deal with, and applications that used to be almost intractable are now suddenly becoming possible. That’s why there’s a lot of excitement and investment as robots start to enter unstructured spaces.
You see thinks like humanoids doing backflips, and the reality is that it’s a bit easier to do those things, even though they look sophisticated, because the robot is controlling its own body. But if you take the example of picking up a complex object with a multi-fingered hand, you have multiple contacts and slippage.
We have nothing like the sensitive and self-healing skin of the human hand. When robots can manipulate objects very robustly and reliably, I think we’re going to find a lot more applications becoming tractable.
What do you think of humanoid robots, humans in the loop, and human-machine collaboration?
Kuffner: I’ve been fascinated by anthropomorphic robots for a long time. They don’t need to fool us into thinking they’re human or enter the uncanny valley. From a practical point of view, if you’re trying to create a general-purpose machine that can solve the largest number of tasks in environments that were built for the human shape, humanoids are a great solution.
But to automate a single task like moving a box in a warehouse — a SymBot can move 25 mph — why make a humanoid with legs? It’s over-engineered for the task. We should always think about what we’re trying to do.
There are tasks where a humanoid shape is more efficient, like dealing with stairs, shelves, door handles, and human-machine interaction, but a lot of tasks don’t have to be constrained by morphology. Other robots can be cheaper, more reliable, or more performant. It’s not an either/or question; the reality is that we’ll have both legged and wheeled robots in the warehouse.
What are your thoughts about physical AI?
Kuffner: When I think about physical AI, what comes to my mind is the conjunction of a digital computer that operates on discrete bits of information as it interfaces with a physical world that is continuous. So operating kinematics and dynamics of continuous forces and continuous movement have this interesting interaction with the digital.
We’re making tremendous progress. Every year, computers get more powerful and= data processing gets more capacity. That means that we’re able to handle more and more tasks that require reasoning with more complex physical models.
I’ve been very passionate in cloud robotics, talking about how accumulating the data to model a phenomenon that you can’t model analytically but you may be able to model statistically, is a superpower. We don’t have to give up model-based robotics. Classical robotics and mechanics can be used to bound the performance and the motion of the machines, but we can suddenly fill in the gaps where physics has struggled.
What’s the role of simulation at Symbotic?
Kuffner: Digital twin technology is important for two reasons. One, it really allows us to augment the physical data that we’re collecting with synthetic data with lots of variants, and we can test our end-to-end our system with many more edge cases because we can inject failure modes that may have only a one in 100 million chance of happening.
So it’s it’s regression testing, testing the the correctness of your software in in an accurate emulation of the whole system, and it’s also data augmentation, creating large data sets that would be very time-consuming, expensive, or even impossible to gather in the real world. We’re of course leveraging that very heavily at Symbotic to make our systems better.
For example, in just a single day, our large warehouse system approximately 100 trillion bits of data across all of our systems and robots. That’s about 16 terabytes a day, and we’re processing that. Each SymBot has over 350 TOPS processing power, so trillions of operations per second. That exceeds the processing power of the the average Tesla car on the road today.
We basically have 20,000 of these supercomputers that collectively can gather and process just incredible amounts of data, and and all of this can be validated and tested with a digital twin. We can create the models, and then when we are confident that they are performant, we can then deploy them to production. So it’s a it’s a positive flywheel of innovation.
The other thing that’s interesting is when the robot is unsure of what to do, it has the option of dialing teleoperators at our headquarters. A a human can see instantly the live video and then help guide it, and then remotely even move the case or slightly alter it and give the robot commands.
And we record all that interaction, so that the next time the system encounters a similar edge case, it first tries to execute a learned autonomous behavior based on the teleoperator feedback. That means over time, it gets better and better and doesn’t have to call for help as often. This is a continuous flywheel of improvement.
Robot suppliers must understand performance tradeoffs
Is there progress getting to five 9s of reliability? What are some of the tradeoffs?
Kuffner: In industrial automation for logistics or manufacturing, both of these customers require very high levels of precision, accuracy, and reliability. There’s parts of the Symbotic system that are over six nines of reliability, so less than one in a million failures.
But if you’re moving 10 million boxes a day, that may not be good enough. So it really depends on what the task is, how sensitive it is to failures, and how fast is it to recover. Teleoperation is really a way for us to do failure recovery or mitigation of failures much more rapidly, efficiently, and cost-effectively. At the same time, it provides valuable training data to improve the autonomy.
For each picking, trailer loading and unloading, or handling pallets, each one requires different levels of reliability, precision, and accuracy., What’s the cost when it does fail, or what’s the time to recover? That’s very important metric to track.
We really try to focus on how to create systems that can deliver value and make the business ROI makes sense. That’s where I think some robotic startups fall into a trap where they fall in love with their technology and ask customers to debug and prove the business case for them.
Has Symbotic been affected by global trade challenges and geopolitical uncertainty?
Kuffner: As we have evolved our fleets of robots, we have worked hard to create a modular system, both in terms of software as well as hardware components. We now have many different variants of our SymBots. We have bots that can handle native cases; we have bots that can handle totes. We have bots that operate in chilled and frozen environments, we have mini bots that handle eaches for our break-pack system, and we have automated fork trucks.
We’re really trying to create the most flexible technology stack that allows a perception module to be run regardless of the morphology of the bot that it’s being run on. We have a planning system that can take flexible constraints based on the customer needs so that we don’t have to rewrite software from scratch just to expand to a new vertical.
It’s really about discipline and architecture and design to create a software-defined robot system for us that allows us to tackle some of the challenges of scaling our business. But there’s still a lot of work to do, and of course our supply chain is is getting more complicated as we expand the number and variety of our of our systems.
Looking ahead, what are your expectations for Symbotic?
Kuffner: We’re on a wonderful path with a great growth trajectory and lots of customer demand. A challenge is really to make sure that we have enough bandwidth to meet our existing commitments as well as grow of the business.
We’re looking at not only improving the robustness and performance of our existing systems, but also leveraging all of those technology pieces. We’ve got new energy-dense Nyobolt technology, which is fast-charging, as well as better software algorithms for fleet management and storage.
We’re integrating end-to-end warehouse automation to provide reliability in the supply chain and beyond. You’re definitely going to see some some exciting announcements from us, so stay tuned.