Autonomous cars are here, but they need to behave more like humans. Dmitri Dolgov, chief technology officer of Waymo, the autonomous car firm that started life as a Google internal project, said in an interview this week that while these cars are on the road already, big changes are coming that will enable them to understand how people, dogs and other factors interact with the road.
“At this point, we don’t need new breakthroughs,” Dolgov said in a conversation with MIT Technology Review’s Gideon Lichfield during the two-day EmTech Digital conference in San Francisco, California. “The self-driving cars are here. This is not a matter of when or if. It is a matter of how fast we can grow and scale deployment of this technology in a responsible manner.”
The project originally started in 2009 and has rapidly moved from strength to strength. It started a pilot taxi project in Phoenix, Arizona to ferry a team of “Early Riders” around in Chrysler Pacifica hybrid minivans. These initial tests saw just 400 participants out of 20,000 applicants hailing rides to commute around their area. Waymo has driven over 10 million miles autonomously as of October 2018, and the California Department of Motor Vehicles granted Waymo the state’s first-ever license to test autonomous vehicles without a safety driver. From these very limited trial runs, Waymo is slowly laying the groundwork to expand further.
“That said, however, I do believe we’ll see some major breakthroughs, particularly in the areas of prediction, simulation, and decision-making,” Dolgov says. “Applying some of the modern research results to those applications will allow you to go that much faster.”
Lichfield showed a video of Waymo’s Phoenix vehicles tackling a number of difficult tasks, like kids crossing the road near a school, dogs running in front, and a woman chasing after her dog by cutting into the road. The video also showed the chaotic nature of San Francisco traffic, a situation that General Motors-owned Cruise Automation was seen tackling earlier this year.
Dolgov explained how the videos show the “full complexity of the driving task that we have to tackle,” illustrating why the car needs a “deep semantic understanding” to navigate the world. It’s not enough to write a series of rules for how to drive: the car needs to understand how people and other factors can behave on the road, as it leads to more predictable movements that help traditional cars predict the autonomous car’s actions.
While a smarter car may make more informed decisions about the roads, sometimes it may be unclear why the car has taken a specific course of action. Systems like Drive.AI show information about decisions to the outside world using screens, and Waymo has found similar benefits with informing passengers.
“You might be wondering, why is the car stopped?” Dolgov said. “Showing that on the screen to the passenger, some information about what the car is seeing, thinking and how it’s making its decision, is something that we find really adds to the positive experience of people that use our cars.”
All this adds up to a smoother, more predictable drive that will help the firm in its plans to expand its taxi service to more states and explore areas like logistics and delivery.