Path Robotics: Teaching the Robot to Weld

For decades, shipbuilding has presented something of a brick wall for industrial automation.

Robots excel at repetition on a large scale; shipbuilding does not.

Walk through an automotive plant and you see the environment where conventional industrial robotics thrives: components arrive at precisely defined locations, dimensions are controlled and the robot executes the same operation thousands of times. Walk through a shipyard and the equation changes dramatically: Steel moves in massive blocks; fit-up varies; weld gaps change; assemblies are measured in tons rather than pounds; and moving the work to the robot can require cranes, transporters and considerable time.

Path Robotics believes advances in artificial intelligence, machine vision, sensing and computing have changed that equation. The Columbus, Ohio-based company doesn't describe itself simply as a robotics company. It calls itself a physical AI company, an important distinction when considering its potential role in shipbuilding.

“Physical AI is different,” said Heather Carroll, Path Robotics’ Chief Revenue Officer. Rather than programming a robot to repeat a predetermined movement, she said, Path combines sensing, vision and intelligence so the machine can “see what is in front of it,” assess the seam, joint and material, develop a weld plan and execute it.

In short, Path is attempting to give the welding robot something traditional automation has largely lacked: the ability to adapt.

You've got these really, really large parts that you cannot easily move around the yard.” A mobile system that can travel to those parts, she added, can change the automation equation.

- Heather Carroll,
Chief Revenue Officer, Path Robotics

Technology Born in Ohio

Path Robotics was founded about a decade ago by brothers Andy and Alex Lonsberry, Ohio natives with deep manufacturing roots. Their father operated a fabrication shop, and the brothers started welding when they were children.

One pursued work in bipedal humanoid robotics, while the other studied computational neural networks. The opportunity they ultimately targeted was considerably less futuristic-looking, but arguably more commercially important: welding.

The founding premise was deceptively simple — could reinforcement learning and AI be used to teach a robot to learn to weld in much the same way a human learns to weld?

That question has driven nearly a decade of development and, critically, data collection.

Today Path operates primarily from a roughly 200,000-sq.-ft. facility in Columbus, OH. Carroll said the company has raised approximately $370 million and employs more than 250 people, with employment expected to reach 350 to 400 by year-end. In 2025, Path quadrupled top-line revenue and surpassed $100 million in bookings, according to Carroll. Its three principal markets are heavy infrastructure, AI/data-center infrastructure and shipbuilding.

That cross-industry experience is important because it has provided the raw material needed to develop Path’s core technology: welding data.

Adaptability is the Difference

Shipbuilding has used automated welding for decades, particularly on panel lines and other repetitive processes. But conventional automation has struggled to penetrate many of the processes that dominate ship construction. Carroll sees two primary reasons.

First, ships and ship components are simply enormous. Moving the component to a fixed robotic cell is often impractical.

Second — and perhaps more important — the real-world component rarely looks exactly like the digital model.

“When you're bringing these two massive pieces together on a ship, it's very gappy,” Carroll said. “It's not laser cut and perfect every time.”

Traditional robots are programmed around an expected geometry. If reality deviates too far from that expectation, the process can fail. Path’s Obsidian AI platform is designed to attack that problem.

Before welding begins, Path's sensing and vision system scans the joint and develops a fill plan based on the customer's welding procedure specification. During welding, however, the system continues collecting information. Carroll said it uses approximately 17 inputs, including amperage, voltage, temperature, torch speed and torch angle, and even incorporates audio sensing. Most significantly, the system can monitor the weld as it is being produced and adjust subsequent passes.

The foundation underneath that capability is data. Path has accumulated more than 10 million inches of weld data from robots operating across multiple industrial applications. That data feeds reinforcement learning used to continually improve the model behind Obsidian.

Think of it as accumulated welding experience — except the “experience” is pooled across machines, applications and millions of inches of completed weld.

That distinction is important because welding data is difficult and expensive to create. Unlike many robotic tasks that can simply be repeated for data collection, once two thick pieces of steel have been welded together, the exercise is finished. The material, energy and time required make real-world welding data costly. Path's years of commercial deployment therefore become part of the technology's competitive moat.

Bringing the Robot to the Ship

The next step could prove particularly interesting for shipbuilders. Most Path installations today employ what the company calls worker cells: fixed systems incorporating a six-axis robotic arm operating within a defined footprint. In April, however, Path launched Rove, its first mobile platform. Rove is a designed to move through a manufacturing facility or shipyard and bring the welding system to the work rather than forcing the work to come to the robot.

For shipbuilding, that reverses one of the fundamental limitations that has constrained robotic welding.

“You've got these really, really large parts that you cannot easily move around the yard,” Carroll said. A mobile system that can travel to those parts, she added, can change the automation equation. Initial production deployments of Rove are planned for late first quarter or early second quarter next year, with Louisiana shipbuilder C&C Marine and Repair serving as a launch partner, according to the interview.

Carroll offered a practical example from a recent shipyard visit. At the end of an automated panel line, she found workers manually completing welds the automated equipment could not reach. The expensive automated line was moving quickly, but downstream manual welding had become the constraint.

Her vision: deploy multiple mobile robotic welders at that point until the backlog disappears, then move those robots elsewhere in the yard.

That gets to the heart of Path's pitch. Automation should not necessarily be applied where it looks most impressive; it should be applied where it removes a production bottleneck.

Into the U.S. Shipbuilding Base

Path's maritime ambitions are already moving beyond demonstrations. Carroll said the company is working with major shipbuilders and commercial yards. Path signed an MOU with HII earlier this year and subsequently was selected as HII's physical-AI partner for welding under its HYPER — High Yield Production Robotics — initiative. It is also working with Louisiana shipbuilders on aluminum vessel construction and complex commercial fabrication.

The HII relationship connects Path directly to another increasingly important piece of the U.S. shipbuilding discussion: distributed shipbuilding.

Rather than performing every fabrication operation inside an already capacity-constrained major yard, foundations, subassemblies and other components potentially can be produced elsewhere and transported to the shipyard for final assembly.

Automation could make that distributed model more viable by allowing fabrication to migrate to locations where specialized welding labor may be scarce.

“We're working very closely with the HII team to collaborate on ways that we can help them increase their throughput and increase production capacity,” Carroll said.

That word — capacity — may ultimately be the most important word in this entire discussion.

Augmentation, Not Replacement

Any discussion of automation inevitably raises the specter of job displacement. In U.S. shipbuilding today, however, the more immediate problem is precisely the opposite: there are not enough skilled workers.

“I've yet to go to one [shipyard] that has said, ‘I've got too many welders,’” Carroll said.

For Path, therefore, the argument is not principally about replacing labor to reduce cost. It is about augmenting scarce skilled labor so yards can increase throughput, potentially operate additional shifts and move experienced welders toward the jobs where human skill remains indispensable.

Carroll said she has yet to see a Path customer lay off welders because automation was installed. Instead, welders move toward higher-value and more complex work, inspection, fit-up preparation and operation of the automated equipment itself. In fact, she said some of Path's best robot operators are experienced welders because they understand the process well enough to recognize problems — sometimes simply by hearing them.

The economic argument follows the same capacity-first logic. Path evaluates existing cycle times and models production running one, two or three shifts with automation operating in parallel. Carroll said customers generally see at least a 35% savings on existing labor spend, though automation makes the most sense where there is sufficient production volume and demand.

More important may be predictability: stable cycle times, reduced exposure to workforce turnover and greater confidence that production schedules can be met.

Rebuilding Shipbuilding

The U.S. maritime industry is awash today in discussion about rebuilding the shipbuilding industrial base. New yards, modernized facilities, workforce development and distributed manufacturing all figure prominently. But simply adding facilities does not solve the workforce equation. That is where physical AI potentially earns a seat at the table.

Carroll argues that the objective is not to automate shipbuilding for automation's sake. It is to use technology where it can unlock constrained production, allowing the skilled workforce already in place to produce more ships.

“If we can build more ships here, we can have more jobs here,” she said, “but we can't do that if the critical constraining skilled labor doesn't exist.”

Welding is Path's starting point, not necessarily its endpoint. The company is already looking toward fit-up, while Carroll sees inspection as another logical application for autonomous AI. Across the wider shipbuilding landscape, sanding, painting and grinding offer additional targets for robotics.

The maritime industry has heard promises about robotics before. What is different in 2026, Carroll argues, is the convergence of rapidly improving AI, machine vision, sensors and computing with a generational surge in demand and a structural shortage of skilled labor.

“The technology is very different even than it was five years ago,” she said. Combined with demand and workforce constraints, “this is different this time, and we have to find a way to make this work.”

For U.S. shipbuilding, that may be the point. The robot does not have to replace the welder. It has to help a shipyard get more welding done — reliably, repeatedly and at scale — with the welders it can actually find.

And if the U.S. intends to materially increase the number of ships it builds, that distinction could prove pivotal.