The former CEO of automotive tech company LeddarTech has launched a physical AI startup to help autonomous vehicle and other robot makers deal with one of the big obstacles to real-world deployment: unpredictable, dangerous edge cases.
Montréal-based Qvyon Technologies, led by co-founder and CEO Frantz Saintellemy, publicly launched on Thursday with $8 million USD ($11.4 million CAD) in seed financing. The round, which closed in April, consists of $6 million USD in equity from Mac Venture Capital, BDC Capital’s Black Entrepreneurs Fund, BKR Capital, and Spotlight VP, as well as $2 million in venture debt from RBCx.
“We see an opportunity for us to scale very rapidly and to create a foundational company here in Canada.”
Frantz Saintellemy, Qvyon
Qvyon’s goal is to allow companies to test and validate AI-enabled robots and self-driving vehicles before putting them to work. As generative AI capabilities have ramped up interest in deploying smart robots in factories and on roads, the unpredictability of the physical world (and random human behaviours) has created challenges for companies hoping to earn public trust and drive adoption.
Saintellemy told BetaKit that when he was working on autonomous vehicle software at LeddarTech, the biggest challenges were finding enough real-world data to reliably test and validate the systems.
Founded in 2007 as a spin-off of Québec’s National Optics Institute, it developed environmental-sensing solutions for autonomous vehicles and driver-assistance systems. After going public on the Nasdaq in 2023, the company faced challenges repaying its debts, eventually cutting nearly all its staff and filing for bankruptcy last year. Some of Qvyon’s other founding members are from LeddarTech as well.
Now, Saintellemy and the team are taking the lessons learned beyond the automotive sector to help companies validate their robots across all kinds of hazardous situations.
“These situations were difficult to anticipate, expensive to capture, sometimes dangerous to reproduce, and sometimes statistically impossible to recreate in nature,” Saintellemy said in an interview. “But these are the types of scenarios any autonomous system must learn to handle.”
Simulation and synthetic data alone couldn’t solve the problem, he said, so Qvyon took a different approach. Its software product, MIMIC Core, blends real-world data—which a company can input from its own cameras or sensors—with an AI system that can then generate tons of variations from one recording, allowing a company’s engineers to learn where their systems are making mistakes.
Jason Baibokas, managing director at BDC Capital who leads the Black Entrepreneurship Fund, said in a statement to BetaKit that every industry putting AI into physical action will have to prove it’s safe before deployment. “Qvyon is building the answer,” he said.
RELATED: LeddarTech cuts 95 percent of staff in bid to rescue its business
“Edge cases” with unfortunate outcomes have hurt public perception for companies like self-driving giant Waymo, which has had cars accidentally drive over an exploding firework or be unable to navigate out of gridlock.
Some self-driving companies have developed validation tech in-house. Toronto-based Waabi has its Waabi World AI system to train and improve the performance of its autonomous vehicles on synthetic data, which it says is a cheaper, safer, and more scalable method than traditional road testing-first strategies.
Qvyon says its tech can be a complement to existing AI simulators, like a “reality infrastructure” layer on top that uses real-world data—which Saintellemy is hoping to sell to similar companies in the driving and industrial sectors. One of its first customers is the autonomous trucking company NuPort Robotics, based in Markham, Ont., which partnered with Canadian Tire and says it is testing trucks with customers in multiple provinces.
Beyond the self-driving sector, it has landed another customer in US manufacturer Voltava, which Qvyon says will use its tech to validate inspection systems in factories.
Qvyon’s roughly 20-person team plans to use the funding to deploy the MIMIC platform and land customers in the US, Europe, and Japan, with a roadmap to launching new products in 2027: a system to measure and document test results, and one to improve the AI system using information from machines already in use.
“We see an opportunity for us to scale very rapidly and to create a foundational company here in Canada,” Saintellemy said.
Feature image courtesy Qvyon Technologies.