Driving is hard. Professional truck drivers know this. From “four-wheelers” who don’t understand how to drive safely around 80,000-pound vehicles to the kid who darts out from behind the parked car, the unexpected challenges can test the patience of any driver. Decisions must be made constantly — and instantaneously. One wrong decision and someone can die.
Those challenges are also what developers of autonomous vehicles point to when promoting their technology, promising that roadways will be safer if humans have less control. But as the professional driver already understands, it is nearly impossible to account for all the possible scenarios, and even the most advanced developers have run into scenarios for which they couldn’t design. An example: In 2017, Volvo car engineers in Australia identified a flaw in their self-driving technology that only came to light under real-world conditions.
“We’ve noticed with the kangaroo being in mid-flight … when it’s in the air it actually looks like it’s further away, then it lands and it looks closer,” Volvo Australia technical manager David Pickett said at the time. That caused the self-driving technology to speed up and then rapidly slow down the vehicle. Volvo, of course, has corrected this, but it is an example of a real-world decision that the autonomous car needed to make but it wasn’t smart enough at the time to do so.
So how do you make autonomous systems smarter? The answer may reside in a virtual training school for autonomous systems.
“The simulator can design the test by itself, it can assess the skills of the Waabi Driver and it can learn to drive on its own,” Raquel Urtasun explained to FreightWaves in describing her startup’s Waabi World simulator.
Waabi World is a scalable, closed-loop simulator that Urtasun believes is “the key to unlocking self-driving technology at scale.”
Unlike most simulators that require real-world data gathered over hundreds of millions of miles to be programmed into the system, the Waabi World simulator utilizes artificial intelligence that can generate an infinite number of possibilities the vehicle may encounter. It then uses this endless simulation to quickly train the Waabi Driver — the actual self-driving technology — on how to handle these scenarios when they occur in the real world.
It might even save a kangaroo or two.
“If you want to remove these millions and millions of miles of driving in the real world, then you need to [automate the process],” Urtasun said. “Other sims use human-designed models. It is very expensive to create those worlds [and often look more like video games]. We utilize AI to automatically recognize and create digital twins of everywhere the sensors [have been].”
Urtasun said Waabi, a company she founded in 2021, collects data from real-world driving to create the digital twins. Using AI, Waabi World can add or remove items from the scenarios to create new variations or entirely new scenarios.
As chief scientist for Uber ATG, Urtasun played a firsthand role in developing the technology that powered Uber ATG’s self-driving division’s vehicles, and she has spent decades studying the technology. The division was eventually sold to Aurora in December 2020, but Urtasun moved on, founding Waabi. In June 2021, Waabi emerged from stealth with one of the largest series A funding rounds ever in Canada.
Urtasun said that AI has always taken a back seat in the self-driving technology development approach, something that Waabi is hoping to change. Waabi is a software developer, not a truck developer, that will partner with OEMs and Tier 1 providers. Waabi will focus on the technology side and is utilizing its Waabi World to enable testing at scale for both common driving scenarios as well as unusual situations. This allows it to develop its technology without the need for physical vehicles to spend hundreds of hours on roadways hoping to encounter various scenarios.
“Waabi World is the most scalable, highest-fidelity closed-loop simulator ever and the key to unlocking the potential of self-driving technology,” the company wrote in a blog posting announcing the technology. “Powered by AI, it is an immersive and reactive environment that can design tests, assess skills and teach the self-driving ‘brain’ to learn to drive on its own. We like to call it the ‘ultimate school for self-driving vehicles.’”
Watch: Waabi World explained
Waabi World features four core capabilities.
- It develops digital twins of the world from data, done automatically and at scale.
- It performs near real-time high-fidelity sensor simulation that enables testing of the entire software stack in an immersive and reactive manner.
- It creates scenarios to stress test the Waabi Driver (aka the “brain” of the simulator).
- It teaches the Waabi Driver to learn from its mistakes and master driving skills without human intervention.
Urtasun said what makes Waabi World and Waabi Driver different from other simulators is the ability for the system to learn.
“When self-driving vehicles are testing on the road, they are not learning anything,” Urtasun added. “If they see the same scenario 30 minutes later, they would behave the same way. This is not how humans act in the world. In Waabi World, every time we experience a scenario, we are able to provide feedback to the Waabi Driver to handle the scenario better the next time. The more [experience you generate] the more it learns.
Watch: FreightWaves’ Alan Adler talks autonomous trucking
“It is telling the system what you are doing incorrectly, just like a teacher … but also what you are doing well,” she added.
This ultimately allows Waabi to create an endless possibility of scenarios to help the Waabi Driver to experience possible real-world situations without them having to have occurred in the real world first.
“Think of it like a video game, where every action has a reaction. Specifically, the simulator tells the ‘actors’ in the scenario where to go and what to do. The simulated sensors that see the updated world then tell the Waabi Driver what it would observe, and then the Waabi Driver decides how it will react. The simulator then moves the Waabi Driver in the virtual world according to its decision and the other traffic participants react to it. This loop goes on and on,” the company said.
The other differentiator is that Waabi World is not like traditional simulators that appear like video games, but rather imagery that looks and feels authentic to the real world. This helps better understand and “see” the scenarios.
“Artist-designed worlds often lack the accurate physical properties needed for fully physics-based simulation, which results in unrealistic sensor data,” the company said. “Instead, Waabi World leverages AI along with simplified physics-based rendering to simulate realistic sensor data in near real time. Our AI algorithms, combined with our high-quality recreated virtual worlds, learn to make the physics approximation look more realistic, while being computationally more efficient than traditional complex physics simulators.”
Welcome to Waabi World.