Waymo details ‘Carcraft’ simulation software, ‘Castle’ testing site for self-driving car training

While real-world experience is important for training self-driving cars, virtual driving is increasingly playing a bigger and bigger role for Alphabet’s autonomous vehicle division. This Carcraft virtualization works in tandem with a “secret base” known as Castle where Waymo has built a city that can quickly be reconfigured to different testing scenarios.

In an interview with The Atlantic, Waymo went into depth about their testing regiment for self-driving vehicles. Named after World of Warcraft, Carcraft was originally software for playing back driving experiences. However, it has since become a key tool for running simulations and is now responsible for teaching their system new things.

At any time, there are now 25,000 virtual self-driving cars making their way through fully modeled versions of Austin, Mountain View, and Phoenix, as well as test-track scenarios.

In 2016, they logged 2.5 billion virtual miles versus a little over 3 million miles by Google’s IRL self-driving cars that run on public roads. And crucially, the virtual miles focus on what Waymo people invariably call “interesting” miles in which they might learn something new.

If vehicles encounter a particularly tricky situation and need extra practice, Waymo will replicate the troublesome situation at their Castle site in Central Valley, California.

They signed a lease and started to build out their dream fake city. “We made conscious decisions in designing to make residential streets, expressway-style streets, cul-de-sacs, parking lots, things like that,” she says, “so we’d have a representative concentration of features that we could drive around.”

To date, 20,000 driving scenarios that originated from Castle have made their way into the Carcraft simulation. This notably does not include other scenarios that came from public driving and other sources.

The key advance Waymo thinks it has cracked with its self-driving system is understanding the intent of others vehicles and pedestrians on the roads.

“Our cars see the world. They understand the world. And then for anything that is a dynamic actor in the environment—a car, a pedestrian, a cyclist, a motorcycle—our cars understand intent. It’s not enough to just track a thing through a space. You have to understand what it is doing,” Dmitri Dolgov, Waymo’s software lead, tells me.

After years of testing, Waymo believes that it has learned how to take in live data from the real-world to create that “understanding of the scene” necessary for self-driving vehicles to flourish.