Waymo may be winning the race to build fully autonomous cars (at least by some estimates), but it still has a long way to go before those vehicles can roll themselves into showrooms. Among the more challenging tricks that self-driving cars have to master next is recognizing emergency vehicles and responding appropriately.
That's more complicated than it sounds. Computers are great at lots of things, but pattern recognition isn't one of them. Teaching autonomous vehicles to identify traffic lanes, for example, is relatively easy, so long as a roadway's painted lines are well maintained. Teaching them to distinguish between cats and dogs, on the other hand, takes much more time, energy, and creativity.
Unfortunately for Waymo (and every other company working on self-driving cars), training a computer to spot an ambulance is roughly as challenging as training it to spot an Airedale. And to complicate matters, not only do autonomous cars need to be able to spot such vehicles, they also need to know how to react quickly.
For example, if an ambulance is approaching from behind, a self-driving car needs to respond by pulling over and letting the ambulance pass. If traffic is heavy, it needs to work with other vehicles to move out of the way.
On the other hand, if a police car or fire engine is approaching from the left or right at an intersection, the autonomous car needs to wait for the vehicle to pass--even if the self-driving car has the right of way.
This is what Waymo is in the process of doing. And it's doing it the tough way: by gathering a range of images and impressions of emergency vehicles, then training its autonomous cars to spot them. It's also giving cars the tools to identify emergency vehicles when they don't look or sound like they've been taught--for example, when they're coming from an odd angle or are strangely lit.
Waymo is currently gathering such data in Arizona, where it runs one of the country's largest public trials of self-driving vehicles. The company says that it's made huge improvements, thanks largely to its custom-built hardware and software (a not-so-subtle jab at Waymo's favored enemy, Uber).
Of course, there's still no word on when cars equipped with Waymo's self-driving systems might be available to consumers. That info might become available later this year, when Tesla is expected to launch a fully autonomous version of Autopilot for consumers. A positive reception from the public could encourage Waymo and others to ramp up their distribution timelines. Stay tuned.