Chris Urmson heads up our driverless car program. Prior to joining Google, he was on the faculty of the Robotics Institute at Carnegie Mellon University, where his research focused on motion planning and perception for robotic vehicles. His self-professed motivation for his work? Making sure that his 11-year-old son doesn't get a driver's license in four and a half years.
Q: Why is Google working on self-driving cars?
A: Because we’ve always looked for ways that technology can change the world. More than a million people worldwide die each year in traffic accidents—94% of which are caused by human error. If we can solve this, it will prevent the majority of traffic-related deaths and injuries, and also help the millions of people who are unable to drive because of disabilities.
Q: Where is the program right now?
A: We’ve reached a couple of major milestones this year. We've now self-driven over one million miles total in our Lexus SUVs, and continue to cover 10,000 miles each week—about what a typical American adult drives in a year. Our new prototype vehicles—with safety drivers on board—are now on public roads in Mountain View. We've also got two Lexus SUVs in Austin, Texas, so we can learn from different driving environments, traffic patterns, and road conditions.
A: We actually did start out by modding two existing vehicles, the Toyota Prius and the Lexus RX450h. But designing our own prototype from scratch opened up possibilities that can’t exist in a car that’s built for and around a driver. For example, we were able to take out the steering wheel and pedals, and change the shape of the vehicle so our sensors can be placed for their optimal field of view. We were also able to build in backup systems for braking, steering, computing and more into the vehicle.Q: Why build your own cars instead of using existing vehicles?
A: Automakers are focused on driver assistance systems like advanced cruise control and automated parking. But in those systems, the driver is still expected to take over as needed. With fully self-driving technology, which is what we’re working on, the car is designed to do all the work of driving all the time, and a human is never expected to take control of the vehicle. We think this will have the biggest impact on safety and mobility for people.Q: What’s the difference between your self-driving technology and what automakers are doing with autonomous driving?
Q: There are lots of articles about all the accidents Google’s cars have been involved in. Is that a sign that they’re not safe?
A: In six years, over the course of 1.8 million miles of autonomous and manual driving, we’ve been involved in 15 minor fender-benders. The self-driving car was never the cause. And except for the most recent incident, where some minor whiplash was reported, there haven’t been any injuries. Instead, given that we were rear-ended in 11 of those 15 incidents, the cause seems to be distracted drivers who aren’t watching the road. For anyone who wants the details, we now publish monthly reports with summaries of all incidents.
However, there’s something of a silver lining here: This is the first time we’ve been able to aggregate data about the rates of minor accidents on American city streets. They typically aren’t reported to police because there's little damage and no injuries. And given that we cover as many miles in one week as the average American does in a year, and have yet to cause an accident while in self-driving mode, we think we’re doing pretty well compared to human drivers.
Q: How do you plan to bring the technology to market?
A: We’re still figuring that out. We’re going to learn a lot from our testing in the next few years, including how people might like to use our technology in daily life. If it develops like we hope it will, we’ll work with partners to bring it to users. And maybe the best way to do that will be to use a car-sharing model rather than traditional ownership...but to be honest, it’s way too early to know how that would work.
Q: Sometimes human drivers have to make tough choices really quickly, like whether to speed ahead to avoid a potential collision, or slam on the brakes and risk getting hit. How does a self-driving car know how to react in those kinds of situations?
A: There’s no way to pre-program the car to do a particular thing in every possible situation—there are literally infinite scenarios that a vehicle could encounter, so it’s just not feasible. Instead, our technology gives the car fundamental machine learning capabilities to respond correctly to unexpected situations as they happen. The more our cars drive, the more scenarios they encounter, and the better the technology gets at handling them.
Q: What’s up next?
A: We really want to learn how communities perceive and interact with self-driving vehicles, so if you’re in the San Francisco Bay Area or Austin and want to share your opinion, we’re eager for feedback. We’ve also launched a public art project where artists in California can submit their work to be featured on our cars. And if you want to get the latest updates, check out our website, or follow us on Google+ or YouTube.