Raquel Urtasun Launches New Self-Driving Tech Startup

A New Approach to Autonomous Vehicle Development

A significant question surrounding Uber’s divestiture of its Uber ATG self-driving division to Aurora has now been answered.

Raquel Urtasun, a leading figure in artificial intelligence and former chief scientist at Uber ATG, has founded a new company, Waabi. This venture focuses on accelerating the commercial viability of autonomous vehicles, initially targeting long-haul trucking, through what she terms an “AI-first” methodology.

Securing Investment and Expertise

Urtasun, who serves as the sole founder and CEO, has already garnered substantial support. Notably, both Uber and Aurora have made separate investments in Waabi.

The company recently completed a Series A funding round, raising $83.5 million, led by Khosla Ventures. Additional investors include 8VC, Radical Ventures, OMERS Ventures, BDC, and Aurora Innovation. Furthermore, prominent AI researchers such as Geoffrey Hinton, Fei-Fei Li, Pieter Abbeel, and Sanja Fidler are also backing the startup.

Waabi’s Core Philosophy

Currently employing 40 individuals across locations in Toronto and California, Waabi represents the culmination of Urtasun’s career-long dedication to bringing practical self-driving technology to fruition. The company’s name, meaning “she has vision” in Ojibwe and “simple” in Japanese, reflects its guiding principles and ambitions.

Existing autonomous vehicle companies utilize a combination of artificial intelligence and sensor technology to replicate human driving tasks. This includes perceiving the environment, interpreting objects, and making safe navigation decisions in various conditions.

Addressing Limitations in Traditional AI

While most developers employ traditional AI methods, Urtasun argues that this approach restricts the full potential of the technology. She explains that developers are often required to manually fine-tune the software, a process that is both complex and time-intensive.

Consequently, Urtasun believes that the development of autonomous vehicles has been hindered, and current commercial deployments are limited to simple operational environments due to the high costs and technical challenges associated with scaling.

“Through years of experience in this field, and particularly within the industry for the last four, it became increasingly evident that a novel approach, distinct from the conventional methods employed by most companies today, is necessary,” stated Urtasun, who also holds a professorship at the University of Toronto and co-founded the Vector Institute for AI.

The Role of Deep Neural Networks

Some developers utilize deep neural networks, a sophisticated AI technique enabling computers to learn patterns from data through interconnected networks. However, these networks are typically isolated to specific tasks, integrated with machine learning and rule-based algorithms within the broader system.

Deep neural networks present their own challenges. A key concern is the “black box” effect, where the reasoning behind an AI’s decision-making process remains unclear. This lack of transparency poses a problem for verifying and validating self-driving systems.

Furthermore, it’s difficult to integrate existing knowledge about the task – such as the fundamentals of driving – and deep nets demand vast amounts of data for effective learning.

Waabi’s Innovative Solution

Urtasun asserts she has overcome these limitations by integrating deep neural networks with probabilistic inference and complex optimization algorithms. This combination allows for tracing the AI’s decision-making process and incorporating prior knowledge, reducing the need for extensive retraining.

A closed-loop simulator is also a key component, enabling the Waabi team to test a wide range of driving scenarios, including critical safety situations, at scale.

Balancing Simulation and Real-World Testing

Waabi will maintain a fleet of physical vehicles for on-road testing. However, the simulator will significantly reduce the reliance on this method. “We can even prepare for new geographic locations before physically deploying there,” Urtasun explained. “This offers a substantial advantage in terms of scalability.”

Collaboration, Not Disruption

Urtasun’s intention is not to disrupt the existing ecosystem of OEMs, hardware, and compute suppliers, but rather to collaborate within it. This collaborative approach may explain the investment from Aurora, a company developing its own self-driving stack for logistics applications like long-haul trucking.

“This was the opportune moment to pursue a different path,” Urtasun concluded. “The industry requires a diverse range of approaches to achieve success, and it became clear that this was the optimal direction to take.”