Waabi CEO talks using AI to train, scale autonomous trucks

Self-driving trucking startup Waabi raised $200 million in series B funding from corporate investors like Uber (UBER), Nvidia (NVDA), Volvo, and Khosla Ventures. Waabi aims to leverage generative AI for practical uses of autonomous driving, especially for commercial-grade trucks.

Waabi Founder and CEO Raquel Urtasun sits down with Yahoo Finance to explain how the company is in the process of training autonomous vehicles through AI models and simulations, commenting on the progress of autonomous driving over last several decades.

"So Waabi has created a new generation of foundational models that are able to understand and interpret what they see, be able to create abstractions that are interpretable so that you can validate and verify the system," Urtasun tells Julie Hyman on Asking for a Trend, "and then uses those abstractions to reason something that for the first time foundational models could not do before, and this gives the system, the ability to really generalize to the unknown, which is fundamental when you bring you want to bring this type of technology to the physical world."

For more expert insight and the latest market action, click here to watch this full episode of Asking for a Trend.

This post was written by Luke Carberry Mogan.

Catch up on Yahoo Finance's latest reporting on autonomous and full-self driving developments:

Investors are buying into AI and Musk any way they can: Morning Brief

What ridesharers can unlock in autonomous vehicle partnerships

Tesla's robotaxi unveiling won't just be about the technology: 'We're starting a new book'

Alphabet's Waymo faces probe from US regulators

Tesla is in 'incredibly enviable' position for real-world AI

Safety is 'core of everything' for Waymo's robotaxi service

Cities are 'the biggest hurdle' for Tesla's robotaxi: Analyst

Video transcript

It's been a bumpy road for many self driving vehicle start ups, but Wabi approaches the tech differently.

The autonomous vehicle start up is deploying generative A I in its quest to launch driverless trucks by 2025.

And Big tech is taking notice earlier this week, the company raised $200 million from investors including Uber, NVIDIA and Volvo's Venture Wing.

Joining me now the CEO of Wabi Raquel orta San uh Raquel.

Thank you so much for being here.

So, first of all, congratulations on the race.

Talk to us a little bit about uh what Wabi does and how you're differentiating from some of the other autonomous vehicle companies out there.

Yeah, thanks Julie for having me here.

Um Yes.

So if you look at the industry today, uh particularly that I've been tracking the industry has consolidated to a very hand in year approach, what we call it V one point.

Oh, that really has difficulty solving.

The problem is very capital intensive.

And what um what he is doing is bringing the revolution of United A I to the physical world.

And uh that's what uh you know, give us a very significant advantage in terms of how quickly we can actually build and deploy this technology as well as uh you know, being super capitalistic compared to the rest of the competition.

And so how does the, the training um exactly work of the software uh that you're using to, to um get to autonomy?

Yeah.

So what has created a new generation of foundation models uh that uh are able to uh understand and interpret what they see, be able to create abstractions that are interpretable so that you can validate and verify the system and then uses those abstractions to reason something that for the first time, you know, foundational models could not do before.

Um And this gives uh you know, the system, the abilities to really generalize to the unknown, which is fundamental when you bring you gonna bring this type of technology to the physical world, so that it is very data efficient as well.

It's very efficient in terms of computation on the vehicle and it's provably safe.

Uh You can think of it as we solve the alignment problem uh that you see uh in traditional more transformer like architectures uh that you see on the last language model side of things.

Um And the way that we train the system is mostly on simulation where we create kind of next generation also genetic P I technology so that we can expose the system to all the 60 critical situations including accidents, et cetera and when we put the system in the real world, it's already much more performed than anything that you see out there.

Uh We, we're just showing Wy trucks on the road.

Do you guys?

But do you guys don't make the trucks?

Right?

You are making the software that goes into the trucks.

How does that sort of melding work?

Yeah.

So we integrate uh our sensors and computes.

So we don't build the sensors themselves, but we actually patch them in a way that is very uh uh you know, build so that it's very easy to integrate at the factory line with the OEM.

Uh So we partner with the O EMS and then uh you know, our technology can be very robustly uh deployed in the physical world.

And then the, you know, the biggest I will say kind of a differentiator is the software uh that drives the trucks is uh you know, for the first time is genetic P I uh that can really do uh you know, with an end to end system, as I mentioned, that is interpretable, right?

Something that is kind of like the next frontier uh for the industry uh that has uh you know, really give us this competitive advantage compared to anybody else.

Well, um we've been talking about autonomous for a while now, you know, we've heard people like Elon Musk promise it for a while.

And yes, as I mentioned, we're supposed to be getting this announcement later this summer, we'll see exactly what that announcement entails.

Why is autonomous?

Why is it so hard to crack?

And are we really closer, not just at Wabi, but at other entities?

Are we really getting closer to cracking it?

Yeah.

And as for context, I spend almost two decades now working in, in Saudi Arabia.

Right?

And, and you know, the um you know, the problem has been really that um you know, the approaches that the industry had utilized to solve self driving were, were very fine engineered were actually not scalable.

Um that really had difficulty uh generalizing to the unknown.

And there are so many things that can happen in the physical world as you drive that uh you know, the systems were, you know, really uh you know, are having uh you know, more and more difficulties, they make more, you know, to make progress towards so that, you know, it's been a technology problem uh so far.

And, you know, the reason I started the company three years ago is that being at the forefront of innovation and really saw that the industry needed a different approach that, that uh kind of uh approach that will be scalable, will come from an A I first approach.

Uh and really ambition the full power of A I uh to build a single A I system that can actually really, so this task and what we see today is that, you know, three years in uh this technology is so powerful and you've seen the revolution, I would say in the digital world with a GP T in November 22.

And uh you know, all the, you know, different things that we have seen from other companies afterwards.

Uh And you know, the same kind of, you know, the cas of a revolution in the physical world.

And this is not possible prior to this technology.

Uh So definitely what you're gonna see now is uh you know why we deploying driverless next year, uh our commerce product.

And the other thing that we uh we have also for the first time in public is that uh we're gonna do many more things just tracking.

Uh We're gonna do other, other robots with different form factors, whether it's robot access, whether it is warehouse robotics, uh whether it's humanoids uh will be things that, you know, we will be doing in the future and all thanks to this, you know, groundbreaking technology.

Well, we will keep in touch as you make all those developments for CAL.

Thank you very much.