Automated welding keeps running into one stubborn physics problem. Heat warps the joint as the torch passes, and the part shifts again on cooldown, so a robot locked to a preprogrammed path aims at a target that has already moved.
Path Robotics Inc. has raised $300M in a Series E at a $2.45B post-money valuation. Georgian led, with Yamaha Motor’s venture arm and Navier, a large South Korean tech company, among the participants. The Columbus, Ohio company announced the round Tuesday.
A model that watches metal move
Path’s answer is Obsidian, a model built for torch-equipped industrial robots. It reads how a part deforms mid-weld and reshapes the workflow in real time, absorbing the small differences between parts stamped from the same blueprint.
Four laser emitters mounted on the torch light up the seam, and two cameras capture the bounce-back. Path says Obsidian collects signal before, during and after every run, feeding a weld world model that researchers then use to generate simulated jobs and train the controller.
Obsidian began on large fixed welding cells. Rove, a four-legged platform the company showed off in April, carries the same software into shipbuilding. Vehicle components, generators and utility poles are further targets on Path’s list.
Capacity, not novelty, is the constraint
Chief executive Andy Lonsberry frames the raise around output, not replacement. Manufacturers need more production than conventional automation can supply, and many simply cannot add lines.
The money scales the welding business alongside customers and funds the data, compute and model work required to reach tasks upstream and downstream of the weld itself.
Physical AI is now one of the more heavily funded corners of industrial tech. The question investors are underwriting is whether a learned model holds tolerance in a shipyard the way it does in a demo cell.
