[
Mobile robots currently move only a small percentage of goods in U.S. warehouses. While many companies are developing physical AI, Atlas Robotics said it has already been deploying the technology in commercial operations.
Material handling in the U.S. is a more than $100 billion labor market, and most of the 2.5 million pallet movements every day are still done with manual forklifts and pallet jacks, noted Atlas Robotics. For more than a decade, automated guided vehicles (AGVs) and autonomous mobile robots (AMRs) have spread slowly, and physical AI promises to accelerate adoption.
“We started Atlas Robotics because the industry kept settling, building machines that needed the world to bend around them instead of robots that handle the world as it is,” stated Çetin Meriçli, co-founder and CEO of Atlas Robotics. He and brother Tekin Meriçli founded Atlas Robotics in 2017.
The Pittsburgh-based company said its mission is to develop systems following its “one brain, many bodies” principle. Atlas Robotics said its vertically integrated AMRs, fleet management software, and fleet-wide learning loop can turn “any brownfield warehouse into a best-in-class automation operation with no infrastructure or processes changes required.”
Nic Temple comes aboard as CRO
Atlas Robotics has named Nic Temple as chief revenue officer as it scales its platform. He has experience with both traditional and autonomous forklifts.
“I spent over 20 years in industrial automation, material handling, and robotics. I was one of the first salespeople for Mobile Industrial Robots here in the States, building its office in San Diego, its distribution channel, and its Latin American presence,” Temple told Automated Warehouse. “I worked at Hyster-Yale, where I was exposed to Balyo and had responsibility for the deployment team. I also worked at other companies and VCs.”
Source: LinkedIn
“I had fallen out of love with the automation industry until I met Çetin,” he acknowledged. “I was frustrated that companies didn’t have what it would take to deliver useful warehouse technology for the warehouse floor.”
While doing consulting and market research, Temple said he saw well-funded companies fall short on systems that could meet or exceed human productivity and provide a return on investment (ROI) despite the cost of onboard compute, sensing, and U.S.-based assembly. Temple also wanted automated lift trucks to provide a foundation for the sensing, manipulation, simulation, learning, and decision making needed for physical AI to work in the real world.
“We got this intro e-mail to Nick from an investor. He was trying to understand the business and was very impressed by the technology but was trying to figure out the commercial side of things — dependability, actual value proposition, etc.,” said Mericli. “We brought him on board to help with the diligence, but it’s very rare to have someone with such a deep history and breadth of understanding, down to the types of customers. From a distance, they may look like they are running three shifts, but they only do unloading in one shift, and in the night shift, it’s usually replenishment. He got the level of obsession that we carry about understanding the customer’s business.”
Less than a month after their first call, Temple had visited Atlas Robotics and agreed to join the company. Atlas Robotics said Temple’s appointment marks its shift from proving its technology to scaling it.
Atlas Robotics builds common core for physical AI
Atlas Robotics said its differentiator is a single core AI that extends across a growing family of machines. The company’s flagship product, LeVO, is an AI-operated pallet jack that handles horizontal pallet movement for inbound, outbound, and cross-docking.
LeVO can read barcodes in any position including double-stacked pallets, and it can work in tight aisles and on high-traffic floors around the clock. It is now available.
manipulator with its core AI.
Source: Atlas Robotics
The same core AI runs Mantis, a patent-pending pallet jack with two dexterous arms built for case picking, and LeVO 3D, which adds vertical reach for racking and conveyors. “Vertical integration and a growing patent portfolio widen the moat as the platform expands from horizontal movement into case-pick and vertical handling,” asserted Atlas Robotics.
“We separate physical AI from traditional industrial automation by the level of intelligence capable of understanding uncertainty, noise, and the surprises of the real world,” explained Mericli. “It has the ability to understand nuance and the agency to act that can’t be accomplished by purely scripted operating procedures or rigid workflows. Most robots won’t work unless the world is very predictable, such as with PLCs and welding robots. For the use cases we’re going after in logistics, exception is the rule.”
He added that a mix of AI models that provide flexibility with more formulaic algorithms will enable robots to move products quickly and safely through the warehouse without hallucinating. “Physical AI cannot mess with the real world; it is really very unforgiving,” said Mericli.
Use cases let Atlas bootstrap warehouse comprehension
Since August 2024, a Fortune 500 warehouse operation has run an Atlas fleet for two shifts per day with 99%+ uptime and a physical intervention rate of less than 1%. Over that time, Atlas said the fleet improved its own cycle time by more than 20% through its fleet-wide learning loop, getting measurably faster the longer it ran.
“A big conundrum is to solve the bootstrap problem, because to truly understand the real world, you need to be in it,” said Mericli. “Apart from pilots and ‘innovation theater,’ how do you earn your place in the warehouse in order to learn from it? Our goal from Day 1 was competence. Warehouses are very pragmatic places. We initially relied on more deterministic models to get something good enough to start with and built an architecture that would get better every day. That’s one reason we started with an automated pallet jack, which is a minimum viable product for mobile manipulation.”
“It’s like when a zebra is born — it can immediately start walking,” he said. “These days, people are talking about tool-using agents, because at some point, it was just a big blob of neural networks, and all you need is more data, and eventually they will do everything. Our overall approach to AI is leaning on the symbolic parts for scaffolding to evolving into a tool-using agent as your overall understanding of the state of the world gets deeper and more competent. You still have access to the safe and deterministic perception.”
While every warehouse is unique, solving one customer’s problems with a particular suite of sensors and software makes automation difficult to replicate or scale, added Temple. With a semantic understanding of the environment, Atlas’ AI opens up the potential for hardware to support different workflows, overcoming gaps in integration and unlocking ROI for customers, he said.