Toronto-based autonomous truck technology maker Waabi announced a strategic partnership on Tuesday with Volvo Autonomous Solutions to jointly develop and deploy autonomous trucks. The partnership is twofold, combining Waabi’s innovations in generative AI with Volvo’s autonomous truck, the Volvo VNL Autonomous.
The vertical integration comes in the form of integrating Waabi’s virtual driver system, the Waabi Driver, into the Volvo autonomous truck, which has redundant systems for safe autonomous operations.
FreightWaves spoke with Raquel Urtasun, founder and CEO of Waabi, about the partnership. The first thing to note is this is not a retrofit, but a vertical integration: These trucks will come off the factory line fully equipped and ready.
“We don’t believe that retrofit is an option for that redundant platform; we believe that our technology, and any AV player for that matter, should be vertically integrated into a redundant platform that is purposefully built for self-driving,” said Urtasun.
While Volvo handles the tractor, Waabi has been cooking up its Waabi Driver, part of the next generation of autonomous vehicle progression called AV 2.0.
AV 2.0 and autonomous trucking’s Kitty Hawk moment
The best way to understand AV 2.0 and what that means for autonomous trucking is to first understand what AV 1.0 did.
“If you look at the industry, there was a generation of technology, what we call AV 1.0, which is a more traditional approach that is some engineered systems where there is AI on them but AI plays a very secondary role,” said Urtasun.
She adds that one way to imagine AV 1.0 is smaller AI models all over the place, but there’s a system managed by humans requiring large teams to develop, necessitating many miles to see the next set of edge cases that the system can’t handle. More miles, that is followed by building a bigger dataset, but it’s very capital-intensive and time-consuming, resembling a tedious iteration.
The past two years brought about the next evolution, AV 2.0: “this idea of you can have a single AI system that can do all the tests necessary for driving,” adds Urtasun. A parallel that she notes is to imagine having a large language model (LLM) that can engineer a system, versus having to manually program it in the past.
With great power comes great computing responsibility. One reason Nvidia has recently been in trucking news is that its computer hardware is powering the AV 2.0 renaissance. Nvidia Drive Thor, built on the recently released Blackwell architecture, was designed for transformer, LLM and generative AI workloads. Nvidia Drive Thor also powers Waabi Driver.
One downside to AV 2.0 is that while it can fully power the AI, compared to AV 1.0, it’s considered a “black box” approach. Urtasun adds that with a sole focus on AV 2.0, “It’s very hard [to] verify that the systems are going to do the right thing and they require massive amounts of data so you went from the brute force by humans to the brute force by data chips and data centers, it’s a massive cost.”
This also explains why the recent gains by Deepseek made headlines, as generative AI was thought of as an arms race between computing power and stuff needing to be computed. The workaround to the endless amount of data was to create within the AI the ability to reason and combine the best of both worlds.
Waabi World, Waabi Driver and an AI mind
AI development in autonomous driving is now more closely resembling creating a virtual human brain due to these two challenges: a near limitless need of human teams to manually program AV 1.0 or the near limitless computation needs required to process information from AV 2.0 and feed its hungry black box. This is a very simplistic analogy but a recurring challenge when trying to train AI models.
To tackle this, the next logical step was to create an AI brain to train self-driving vehicles. To train that brain, on-the-road testing isn’t enough, as it would take thousands of self-driving trucks millions of miles to collect the information to cover every possible circumstance.
The self-driving brain is the Waabi Driver, but to efficiently train it, it needs a parallel high-fidelity closed-loop simulator powered by generative AI. This Waabi World has four core capabilities that save the effort of manually testing and collecting data, which still happens through the Waabi Driver.
Waabi World builds digital twins of the world from data, automatically and at scale while performing a near-real-time high-fidelity sensor simulation. This allows the software to be tested in an immersive and reactive manner through stress tests on the Waabi Driver. The Waabi Driver then learns from its mistakes in the virtual world and applies that to mastering the art of driving without a person.
With the combination of virtual training and teaching, sometimes magic happens. Urtasun gave an example of how the Waabi Driver encountered rain, an enemy of autonomous vehicles. “The first time that we went on the road and it was raining, the system wasn’t trained with any rain. It had never seen it before. It still drove perfectly.”
An added benefit of creating a neural simulator with your AI is that they reinforce each other. Urtasun adds, “These two assistants play games in terms of one is the teacher the other one is the student – it gets better and better and better. That’s been also a massive unlock for us because we don’t need to wait for nature to show us.”
Looking ahead
While Waabi is innovating in the AV technology space, the press release notes, “The two companies have laid the groundwork for the integration of the Waabi Driver into the Volvo VNL Autonomous and are preparing for testing in 2025.”
Volvo’s collaboration is part of a yearslong effort where Volvo Group Venture Capital became a strategic investor in the company back in January 2023, then later invested during the company’s $200 million Series B round.
The post Waabi and Volvo Autonomous Solutions announce partnership appeared first on FreightWaves.
