The Fuse

Waymo’s Fake City Shows How AV Testing Is Quickly Evolving

by Matt Piotrowski | September 01, 2017

In order for self-driving cars to make their way onto public roads, the necessary testing needs to be done by stakeholders to continue to develop and improve the technology and convince consumers and regulators, as well as themselves, that autonomous vehicles are as safe as those with a driver behind the wheel.

Performing the right amount of testing to make sure the car doesn’t cause an accident and provides a smooth ride will be daunting and will take some time, but recent news of Waymo’s secret city that it calls Castle shows huge promise on this front.

A recent story in The Atlantic does a deep-dive into Waymo’s Castle city, which is about 100 miles east of Silicon Valley and was built for the sole purpose of testing autonomous cars. The work Waymo is performing at Castle will allow it to combine real-world testing, customizable test-tracking, and virtual simulations to see how they work together.

The work Waymo is performing at Castle will allow it to combine real-world testing, customizable test-tracking, and virtual simulations to see how they work together.

“The miles they drive on regular roads show them areas where they need extra practice,” Alexis Madrigal wrote in The Atlantic. “They carve the spaces they need into the earth at Castle, which lets them run thousands of different scenarios in situ. And in both kinds of real-world testing, their cars capture enough data to create full digital recreations at any point in the future. In that virtual space, they can unhitch from the limits of real life and create thousands of variations of any single scenario, and then run a digital car through all of them. As the driving software improves, it’s downloaded back into the physical cars, which can drive more and harder miles, and the loop begins again.”

What this research shows is that the means to test are innovating and evolving as quickly as the technology itself. Perhaps the most eye-opening and the most groundbreaking is the virtual simulations. The huge benefits of virtual simulations are that they help adjust for weaknesses of real-world trials and experiments, where not every single scenario can be tested.

Simulations could be just as effective, if not more so, than any on-road testing.

Simulations could be just as effective, if not more so, than any on-road testing. These simulations are possible today because of the investments Google has made in high-definition mapping. Cognata, an Israel-based start-up, is another company that has moved forward with AV testing using virtual simulation tools. The company, which has received a lot of positive attention as of late, recreates cities in 3D and uses algorithms and computer vision to test cars, allowing it to drive millions of miles an hour.

As The Atlantic pointed out: “The power is that they mirror the real world in the ways that are significant to the self-driving car and allow it to get billions more miles than physical testing would allow. For the driving software running the simulation, it is not like making decisions out there in the real world. It is the same as making decisions out there in the real world.”

News of Waymo’s testing city reinforces how serious the company is about AVs, and how it wants to be the leader in this space. Earlier this year, it began testing a large fleet and it announced a partnership with ride-hailing company Lyft.

When AVs are eventually on the road at a mass scale, virtual testing will likely be seen as a major process in paving the way for it to happen. While Waymo logged more than 3 million miles of testing on actual roads last year, it was able to record a massive 2.5 billion miles on its virtual Carcraft during the same time period, as it had about 25,000 virtual AVs running all the time. And in virtual testing, vehicles are in special situations with different variables, not just driving on open roads without traffic.

In virtual testing, vehicles are in special situations with different variables, not just driving on open roads without traffic.

The problem with real-world testing is that it is time-consuming and costly, and most areas where AV driving has occurred have ideal weather conditions and don’t have rough driving environments. The Rand Corp. said that AVs will need to drive more than 10 billion miles to demonstrate enough confidence that the failure rate is less than that with humans behind the wheel. This would take years to achieve, making virtual simulations important in speeding up the testing process. “Every autonomous vehicle developer faces the same challenge—it is really hard to generate the numerous edge cases and the wide variety of real-world environments,” said Danny Atsmon, the CEO of Cognata.

News of Waymo’s intense testing efforts comes at the same time policymakers on Capitol Hill are moving forward with AV legislation.

News of Waymo’s intense testing efforts comes at the same time policymakers on Capitol Hill are moving forward with AV legislation and the Department of Transportation is poised to soon update last year’s guidance released by the Obama administration. The House bill that made it out of committee last month—and will be debated on the floor next week—would give automakers exemptions from existing safety standards, allowing them circumvent red tape and put as many as 100,000 AVs on the road

It’s important that for AVs to share the road with traditional vehicles, stakeholders need to make sure that technology is prepared to deal with myriad possibilities that could occur. Rolling out AVs in a safe manner is crucial to gain the public trust. This is what makes Waymo’s Castle city, where self-driving cars are tested over and over, both in real-world conditions and virtually, so vital in speeding up testing to allow level 4 autonomy to hit the road.

ADD A COMMENT