The problem is that for an AI to learn to deal with the chaos of real-world roads, it needs to be exposed to the full range of events it might encounter. That’s why manufacturers of driverless cars have spent the last decade clocking millions of miles on streets around the world. A few, like Cruise and Waymo, have started testing human-less vehicles in a handful of quiet urban environments in the United States. But progress is still slow. “Why haven’t we seen an expansion of these little pilots? Why aren’t these vehicles everywhere? asks Urtasun.
Urtasun makes bold claims for the head of a company that not only hasn’t tested its technology on the road, but doesn’t even have vehicles. But by avoiding most of the costs of testing the software on real streets, it hopes to build an AI driver faster and more cheaply than its competitors, giving the entire industry a much-needed boost.
Waabi isn’t the first company to develop realistic virtual worlds for testing self-driving software. In recent years, simulation has become a mainstay for driverless car companies. But the question is whether simulation alone will be enough to help the industry overcome the remaining technical hurdles that have prevented it from becoming a viable proposition. “No one has built the matrix for self-driving cars yet,” says Jesse Levinson, co-founder and CTO of Zoox, a self-driving vehicle startup acquired by Amazon in 2020.
In fact, almost every autonomous vehicle company now uses simulation in one form or another. This speeds up testing, exposes the AI to a wider range of scenarios than it would see on real roads, and it cuts costs. But most companies combine simulation with real-world testing, typically going back and forth between real and virtual roads.
Waabi World takes the use of simulation to another level. The world itself is generated and controlled by the AI, which acts as both driving instructor and stage manager, identifying the weaknesses of the AI driver, then rearranging the virtual environment to test them. Waabi World teaches multiple AI pilots different abilities at the same time before combining them into a single skill set. Everything happens without interruption and without human intervention, explains Urtasun.
Driverless car companies are using simulation to help them test how neural networks controlling vehicles handle rare events – a bike courier driving past, a sky-colored truck blocking the way, or a chicken driving through. the road – then modify them accordingly.
“When you have an event that happens infrequently, it takes thousands of road miles to test it properly,” says Sid Gandhi, who works in simulation at Cruise, a company that has started testing fully autonomous vehicles on a limited number of roads in San Francisco. . Indeed, rare or long-tail events may only occur once in a thousand. “As we work on solving the long tail, we rely less and less on real-world testing,” he says.