Inside the Montreal lab training robots to work on real factory floors

Key points

• Vention opened its Physical AI Lab in Montreal in 2026, training AI models on data from real factory floors.
• The company has deployed more than 28,000 machines across a community of over 6,000 factories, including 90 Fortune 500 firms.
• Physical AI revenue grew 400% over the past year, according to Vention.
• GRIIP, a modular pipeline for robotic picking and assembly, launched in February 2026, with an open-source release planned.
• Dr. Joelle Pineau, chief AI officer at Cohere, joined as external technical advisor.

A robot that works perfectly in a lab can still fail on a real factory floor. Noise, variation, unexpected parts: the real world resists every controlled test. Vention, a Montreal-based automation company, thinks the answer is to close that gap by doing the research inside the factory itself.

The company opened its Physical AI Lab in Montreal this year, as first reported by The Robot Report. Training happens on data from actual production lines.

What does the lab actually do?

Real industrial data feeds directly into AI model training; improved robot behaviour goes straight back to factory floors. Founder and CEO Etienne Lacroix told The Robot Report: "Any physical AI foundation model is data-hungry." With more than 28,000 machines already running in client factories, Vention pulls a continuous stream of real-world robot movement data that most research teams can't touch.

Two weeks before this lab opened, we reported on Inbolt's argument that the physical AI bottleneck isn't data at all, but deployment. Vention is betting both matter.

The lab tackles tasks that sound simple but defeat most machines. Kitting, assembling components from multiple suppliers into a precise set for one workstation, is the headline use case. Lacroix described an automotive version to The Robot Report: headlamps, a wiring harness, a mounting bracket bundled in the right configuration so Position 7 on the line has what it needs for each specific car. Getting a robot to do that reliably, across thousands of variations, is still an open research problem.

Dr. Jimmy Li, a McGill University robotics researcher who leads the lab, described the setup: "Our clients aren't waiting for a finished product to test; they're in the room while we build it."

What is GRIIP, and why does it matter?

GRIIP, short for Generalized Robotic Industrial Intelligence Pipeline, is a modular chain of software steps that guides a robot arm through spotting an object, identifying it, calculating the grip and moving without collision. Launched in February 2026, it combines large AI foundation models from partners including NVIDIA with Vention's own proprietary models. An open-source SDK is planned.

Vention is already working with a large automotive manufacturer on high-complexity final-assembly tasks, alongside electronics contract manufacturers handling components from dozens of vendors arriving in different packaging.

What does this mean for manufacturers?

For factory owners, the pitch is robots that handle more variety without a specialist reprogramming them every time something changes. Vention expects deployment costs to fall as models mature, putting the technology within reach of smaller manufacturers priced out today.

What I'd watch: Vention's real advantage isn't the lab itself, it's the 28,000 machines already on factory floors generating data nobody else has licensed. Whether that moat holds depends on how fast rivals like Skild or Physical Intelligence, both named by Lacroix as known quantities, can close the data gap through their own partnerships. The open-source GRIIP release will be the first real test of whether Vention's research leads or just follows.