WAABI
Here’s how that works. At any time when its real trucks drive on a highway, Waabi records the whole lot—video, radar, lidar, the state of the driving model itself, and so forth. It could rewind that recording to a certain moment and clone the freeze-frame with all the assorted sensor data intact. It could then drop that freeze-frame into Waabi World and press Play.
The scenario that plays out, during which the virtual truck drives along the identical stretch of road as the actual truck did, should match the actual world almost exactly. Waabi then measures how far the simulation diverges from what actually happened in the actual world.
No simulator is able to recreating the complex interactions of the actual world for too long. So Waabi takes snippets of its timeline every 20 seconds or so. They then run many hundreds of such snippets, exposing the system to many various scenarios, corresponding to lane changes, hard braking, oncoming traffic and more.
Waabi claims that Waabi World is 99.7% accurate. Urtasun explains what which means: “Take into consideration a truck driving on the highway at 30 meters per second,” she says. “When it advances 30 meters, we will predict where the whole lot will probably be inside 10 centimeters.”
Waabi plans to make use of its simulation to show the protection of its system when looking for the go-ahead from regulators to remove humans from its trucks this 12 months. “It’s a vital a part of the evidence,” says Urtasun. “It’s not the one evidence. Now we have the normal Bureau of Motor Vehicles stuff on top of this—all of the standards of the industry. But we would like to push those standards much higher.”
“A 99.7% match in trajectory is a robust result,” says Jamie Shotton, chief scientist on the driverless-car startup Wayve. But he notes that Waabi has not shared any details beyond the blog post announcing the work. “Without technical details, its significance is unclear,” he says.
Shotton says that Wayve favors a mixture of real-world and virtual-world testing. “Our goal shouldn’t be just to duplicate past driving behavior but to create richer, tougher test and training environments that push AV capabilities further,” he says. “That is where real-world testing continues so as to add crucial value, exposing the AV to spontaneous and complicated interactions that simulation alone may not fully replicate.”
Even so, Urtasun believes that Waabi’s approach will probably be essential if the driverless-car industry goes to succeed at scale. “This addresses certainly one of the massive holes that now we have today,” she says. “It is a call to motion by way of, —show me your number. It’s time to be accountable across the whole industry.”
