Q&A session with Google robot-car tester; Stephanie Villegas

It was love at first sight when Google’s Stephanie Villegas caught sight of one of the company’s self-driving cars in 2011. Villegas was a contractor for Google at the time, working in mapping. “I saw the car parked outside the building, and I thought, ‘That’s what I want to work on.’ “

According to Silicon Valley, Villegas, who holds a bachelor’s degree in fine arts from UC Berkeley, soon landed a position driving on the car team and monitoring the vehicles when they drove themselves. “I’ve been here since the Prius,” said Villegas, referring to the series of vehicles Google has used for its cars, from the Toyota Prius to a Lexus SUV to the bubble-like prototypes currently under testing.

Now, Villegas, 29, works as a “structured testing lead,” a job that sees her and her colleagues devising out-of-the-ordinary roadway situations, called “edge cases.” These scenarios teach Google’s autonomous cars to react to hazards that don’t come every minute, or perhaps even every day, but require an appropriate response to avoid catastrophe. The vehicles receive their education on the test track, and via computer simulations, as well as from data gathered during public-road testing. Each learning experience from each car can be spread to the fleet.

Villegas recently spoke with this newspaper. Her comments have been edited for length and clarity.

Q Let’s say you put a self-driving car on a public road. A dog runs in front of a car approaching in the opposing lane and that car slams on its brakes and slides into the lane in front of the autonomous car. The car reacts. What has it learned, and how?

A We can stage something like that at the test facility. You can have the dog run at different speeds and from different locations. You can have a smaller dog, like a Chihuahua, you can have a Great Dane, you can have a German shepherd. You can have them be different colors because those things look different to human eyes and to sensors depending on the time of day. You can have a different-size car approaching. They can be going different speeds, they can be paying different levels of attention. The car then learns how to respond in situations when things might be slightly different.

If you remove the dog from the equation, this now becomes a scenario where an oncoming vehicle has departed its lane and is now in our direct path. Wholly different test, especially if now there are parked vehicles on the sides of the road, or multiple lanes. Knowing simply what a vehicle looks like when it changes its heading into our direction is learned from that test where you have the dog running across the road.

Q Who comes up with the edge case scenarios?

A Ideas can come from things we’ve encountered in the real world. They can also come from our creative test conductors who have extensive experience behind the wheel. Drivers, engineers and managers alike all have insight into how the car interacts with other road users and have great ideas as to how we can improve, so we also welcome everyone on the team to submit their ideas through a form.

Q What are some dangers your cars have confronted?

A A mattress has fallen from the back of a truck. Children running in the road after balls. People (on skateboards) skitching on vehicles, holding on, like Marty McFly in “Back to the Future.” Or coffining — people lie on their backs on skateboards and go between traffic. Imagine someone lying in a coffin, someone assuming the same position on a skateboard and then riding down the street. They ride between vehicles or under vehicles. It’s pretty death-defying. It’s actually pretty fun. I used to do it as a kid in my parents’ driveway. I’m not condoning that behavior. But very fun. Don’t do it in live traffic.

Q Can the vehicle make value judgments? Obviously rear-ending a car when going 10 mph is a better option than driving onto the sidewalk and plowing into 10 people, right?

A That’s the classic philosophical question, something that I’m not sure anyone is really equipped to answer. Our vehicle is amazing in the fact that it has the ability to predict situations like this and avoid entering into situations that can become quite tricky. So our vehicle can see 360 degrees around it for 200 meters, and having that sort of view of the world allows us to monitor things that are interacting with us and around us, like pedestrians, and cyclists, other vehicles. By being able to keep our eyes on multiple things around us as they’re constantly moving and changing, we’re able to take a defensive approach to most situations that human drivers might not be able to.

Q What weather-related issues make driving difficult for the robotic cars?

A Driving in rain makes things look a little different. The roads become slick and shiny to the human eye, and they become slick and shiny to sensors that you might use on a self-driving vehicle. Things like dust can occlude some things, make things look cloudy and force you to slow down as a human driver. We’re really taking advantage of the different weather, the different climates that we experience in the different places that we test. Here in Mountain View, obviously, we have a very moderate climate. But Austin, Texas, has a lot of rain, and it’s very hot and humid. Phoenix is very dry and dusty, and they have huge tumbleweeds running through the desert. It’s also very hot.

Q When will Google autonomous cars hit the mass market?

A Chris Urmson (Google’s head of the self-driving car project), hopes his 12-year-old son never gets his driver’s license. That’s where we’re at.

Advertisements