Today, many vehicles benefit from some form of automated system, be it active cruise control, lane departure warning systems or forward collision warnings. These systems provide greater safety to drivers, can warn of impending dangers, and help prevent crashes. However, a fully autonomous vehicle that does not rely on any form of human involvement remains a distant prospect – but that may be about to change.
Raquel Urtasun, a professor in the Department of Computer Science at the University of Toronto, is the founder and chief executive of Waabi, a new Toronto-based artificial intelligence (AI) company that’s building the next generation of self-driving technology. According to Urtasun, the current problem with many autonomous vehicle startups is that they are often “robotics centric”, requiring humans to provide hard-coded solutions to the vast multiplicity of potential problems a vehicle may encounter on the road. Developers must constantly tweak and check their software stacks – the “brain” of the self-driving system – through real-world testing on public roads and highways. But this approach, Urtasun believes, is too complicated, resource intensive, and risky.
“We need something other than this traditional approach,” she says. “Something that is rooted more in automation – something that can discover solutions on its own.”
This is where Waabi comes in. The company places greater emphasis on AI systems to solve problems. According to Urtasun, although other developers do include AI in their methods, they tend to do so in a way that limits its power. Waabi’s “AI first” approach, however, uses a family of algorithms, including deep neural networks – a sophisticated form of machine learning algorithms – in combination with probabilistic inference, and complex optimisation that allows the AI system to “learn” while still being interoperable; a developer can trace back the decision-making process to see how the AI system is operating.
This “learning” takes place in real time through a high fidelity closed-loop simulator called Waabi World. The simulator exposes the AI system, known as the Waabi Driver, to an immersive and reactive environment where it can hone its driving skills through a vast array of experiences and scenarios ranging from common driving situations to safety-critical cases – essentially all the things that could go wrong on the road. This significantly reduces the need for humans to perform miles and miles of real-world tests, making the process safer, cheaper, and more scalable.
Before launching Waabi, Urtasun was the chief scientist and head of R&D at Uber ATG from 2017 to 2021, when the company sold the unit. Many of her Uber team members have since joined Waabi.
The potential applications for her startup’s technologies are wide-ranging, but Urtasun is focusing the company’s attention on the long-distance trucking industry in the first instance. This is because long-distance motorway driving is less complicated than driving in towns and cities. A robot taxi operating in Toronto, for example, would have to make numerous complex decisions on how to pick up and drop off passengers in various dynamic situations. For trucking, however, Urtasun explains, “the challenges are simpler, and it is easier to build this technology around them”. The hope is that once Waabi has a suitable product on the market, it can then generalise the technology for other applications.
Transportation is an industry fraught with danger – current figures from the US Bureau of Labor Statistics show that transportation incidents account for the highest proportion of workplace fatalities in the country, with “driver/sales worker and truck driver” also listed as the seventh deadliest occupation. Although not differentiated for the haulage sector, this long-term data supports the frequently made claim that truck driving is among the most high-risk professions.
Another reason that Waabi is focusing on the long-distance trucking sector is its connection to the ongoing troubles facing the global supply chain. This situation emerged during the pandemic, and little has changed amid the war in Ukraine and general global uncertainty, but it has been amplified by a continuing shortage of drivers, says Urtasun. “If you look at the possible solutions to this crisis, it isn’t going to be more human drivers because of the chronic shortage – which is only getting worse,” she says. Automation, she believes, is the only long-term solution.
Given that AI forms such an important part of Waabi’s aims, it is no wonder Urtasun has chosen to base the company in Toronto. “Toronto is an incredible technological hub, particularly in relation to AI, as a lot of the global talent is here,” she says. According to Urtasun, the city is now home to all manner of businesses interested in AI development and investment, from startups and scale-ups, to large international companies with their own research labs. “There has been a real boom in the last five or so years, which has changed the landscape of the city,” she says.
Sitting at the centre of all this is the University of Toronto, which has been a consistent international leader in the development of AI systems. In fact, it was the university’s expertise in this field that originally led Urtasun to move to Canada, first as a PhD student and then later as a professor.
“U of T’s world-class array of disciplines and leadership in AI make it the ideal environment for visionaries like Professor Urtasun to take their innovations to the next level,” says Leah Cowen, vice-president, research and innovation, and strategic initiatives. “The ability to collaborate with other top researchers, across a wide spectrum of fields, is unparalleled.”
As a member of the faculty, Urtasun worked with colleagues, such as Turing award-winner Geoffrey Hinton, to co-found the Vector Institute – a world-leading research non-profit based in Toronto, which pushes the boundaries of machine learning and deep learning.
It was the University of Toronto’s collegial atmosphere that made this possible, says Urtasun. “It’s an extremely friendly and encouraging environment [and] an incredible place for research that has helped create the next generation of cutting-edge AI systems that are transforming the industry.”
Meet the extraordinary community that’s pushing the boundaries of what’s possible. utoronto.ca/news
