Waymo may have suspended its on-road testing for self-driving vehicles due to, but that doesn’t mean that it’s not keeping busy in the meantime. See, a big part of making self-driving cars work is training neural networks to recognize hazards in the real world, and while doing that on public roads is important, using simulations is also a big part of it.
Waymo can simulate a century’s worth of on-road testing virtually in just a single 24-hour period, so that’s what it’s doing. In fact, Waymo says it’s simulated 15 billion miles of driving to date. For some context, that’s like making 31,394 round trips to the moon or one trip per day for 86 years. Not bad, huh?
“Waymo’s simulation technology allows us to create completely simulated scenarios using data from what we’ve seen and experienced driving in the real world,” Waymo representatives said in a statement. “In simulation, we can manipulate the environment around our Waymo Driver to gauge how it will respond to various situations like continuous oncoming traffic while making an unprotected turn.”
In short, Waymo puts together a greatest-hits simulation from its 20 million miles of actual on-road test data and runs its software through that for the broadest possible variety of challenging situations, thereby helping it to learn to understand them better.
The impressive thing is that Waymo can do all of this during its parent company Alphabet’s stay-at-home order. Waymo’s employees are able to remotely access the company’s networks, as you’d expect of a big tech company. Still, because simulation software and the visualization software that works with it to output something useful to engineers are so resource-intensive, Waymo has had to come up with web-based versions of some of its most-used tools. That means that the company doesn’t necessarily have to ship its engineers expensive and powerful computer hardware to use at home.