AI pioneer Raquel Urtasun starts a startup for self-driving technology with the support of Khosla, Uber and Aurora
One of the lingering mysteries in Uber’s sale of its self-driving Uber ATG unit to Aurora has been solved.
Raquel Urtasun, the AI pioneer and senior scientist at Uber ATG, has launched a new startup called Waabi that is taking an “AI-first” approach to accelerate the commercial deployment of autonomous vehicles. Transport trucks. Urtasun, the sole founder and CEO, already has a long list of high profile funders, including separate investments from Uber and Aurora. Waabi has raised $ 83.5 million in a Series A round led by Khosla Ventures that brings together Uber, 8VC, Radical Ventures, OMERS Ventures, BDC and Aurora Innovation, as well as leading AI researcher Geoffrey Hinton, Fei- Fei Li, Pieter Abbeel. participate, Sanja Fidler and others.
Urtasun described Waabi, who currently employs 40 people and operates in Toronto and California, as the culmination of her life’s work in bringing commercially viable self-driving technologies to society. The company’s name – Waabi means “she has vision” in Ojibwe and “simple” in Japanese – suggests their approach and ambitions.
Autonomous vehicle startups that exist today use a combination of artificial intelligence algorithms and sensors to manage human tasks while driving, such as recognizing and understanding objects and making decisions based on that information to help a safely navigating a lonely road or a crowded freeway. In addition to these basics, there are a variety of approaches, including within AI.
Most self-driving vehicle developers use a traditional form of AI. However, the traditional approach limits the capabilities of the AI, said Urtasun, adding that dDevelopers have to manually fine-tune the software stack, a complex and time-consuming task. The result, says Urtasun: The development of autonomous vehicles has slowed down and the limited commercial deployments that exist work in small and simple operational areas because it is so costly to scale and Technically sophisticated.
The story goes on
“After so many years in this field, and particularly in the industry over the past four years, it has become increasingly clear that a new approach is needed that is different from the traditional approach most companies are taking today,” said Urtasun, who is also a professor at Department of Computer Science at the University of Toronto and co-founder of the Vector Institute for AI.
Some developers use deep neural networks, a sophisticated form of artificial intelligence algorithms that enable a computer to learn by using a series of connected networks to recognize patterns in data. Typically, however, developers lock down the deep nets to solve a specific problem and use machine learning and rule-based algorithms to blend in with the broader system.
Deep nets have their own problems. A longstanding argument is that they cannot be used reliably in autonomous vehicles, among other things because of the “black box” effect, in which it is not clear how and why the AI solves a certain task. This is a problem for any self-driving startup looking to verify and validate their system. It is also difficult to bring prior knowledge of the task the developer is trying to solve, such as driving a car. After all, deep networks require an immense amount of data in order to learn.
Urtasun says she solved these persistent problems surrounding deep networks by combining them with probabilistic inferences and complex optimization, which she describes as a family of algorithms. In combination, the developer can trace back the decision-making process of the AI system and include previous knowledge so that he does not have to teach the AI system everything from scratch. The final piece is a closed-loop simulator that will enable the Waabi team to test common driving scenarios and safety-critical borderline cases on a large scale.
Waabi will continue to have a physical fleet of vehicles that can be tested on public roads. However, the simulator will allow the company to rely less on this form of testing. “We can even prepare for new regions before we go there,” said Urtasun. “That’s a big advantage in terms of the scaling curve.”
Urtasun’s vision and intent is not to take this approach and disrupt the ecosystem of OEMs, hardware and computer suppliers, but to be an actor in it. This could explain support from Aurora, a startup developing its own self-driving stack that it initially plans to deploy in logistics such as long-haul transportation.
“This was the moment to really do something different,” said Urtasun. “The field needs different approaches to solve this problem and it became very clear that this was the way to go.”