Montreal-based Qvyon Technologies came out of stealth on November 12 with $8 million in seed funding to help robot and autonomous-vehicle makers find dangerous edge cases before machines encounter them in the field. The round, led by MaC Venture Capital with BDC Capital's Black Entrepreneurs Fund, BKR Capital and Spotlight VP, closed earlier this year and is only now being announced publicly.
Qvyon is led by CEO Frantz Saintellemy, the former chief executive of automotive sensor company LeddarTech. His pitch is that physical AI systems keep failing in the same way: they're validated against clean, expected scenarios, then break down the first time a sensor gets blocked, a camera glares out, or conditions shift faster than the model was trained for.
How Mimic Core works
The company's first product, Mimic Core, takes real-world sensor recordings — camera, lidar and radar data pulled directly from a customer's own vehicles or robots — and reconstructs them as 3D simulated environments. From there, engineers can inject variations: a sudden lane change during a rainstorm, a robot arm's vision sensor partially blocked mid-cycle, a forklift encountering an obstacle at an angle it's never seen. The idea is to multiply a limited set of real recordings into a much larger library of plausible failure scenarios, without sending a fleet back out to collect more footage.
Saintellemy says the emphasis on customers' own data, rather than a shared training set, is deliberate: Qvyon keeps cloud customers' recordings separate and says it never uses one client's data to train another's models, a distinction aimed at autonomous-vehicle and industrial robot makers wary of leaking proprietary driving or manipulation behavior to competitors.
Why validation is becoming its own market
Qvyon's launch lands at a moment when humanoid and mobile robot deployments are scaling fast, and so is scrutiny of what happens when they fail. A recent academic study found that degraded or blurry camera feeds pushed AI models toward unsafe robot actions the large majority of the time, underlining exactly the gap Qvyon is targeting. As robots move from pilots to paid production work in warehouses, factories and on roads, buyers increasingly want proof a system has been stress-tested against the scenario that actually hurts someone, not just a demo reel.
Qvyon says it will use the capital to expand into the U.S., Europe and Japan while building out its roadmap: Mimic Validation OS, aimed at documenting test results for customers, is due in the first half of 2027, followed by Mimic Learn, which would feed data from machines already in operation back into model improvement, by the end of 2027, according to SiliconANGLE.