0
Select Articles

Hiding in Plain Sight PUBLIC ACCESS

Google's Autonomous Car Applies Lessons Learned from Driverless Races.

Mechanical Engineering 133(02), 31 (Feb 01, 2011) (1 page) doi:10.1115/1.2011-FEB-3

Abstract

This article presents an overview of Google's autonomous car. There are three components that make Google’s driverless cars go: sensors, software, and Google’s mapping database. Most of these sensors are neatly tucked away in the car’s body rather than mounted, laboratory-style, on a roof rack. The exception is the rotating sensor mounted on the roof. It is a Velodyne high-density LIDAR—light detection and ranging—that combines 64 pulsed lasers into a single unit. The system rotates 10 times per second, capturing 1.3 million points to map the car’s surroundings with centimeter-scale resolution in three dimensions. This lets it detect pavement up to 165-feet ahead or cars and trees within 400 feet. Automotive radars, front and back, provide greater range at lower resolution. A high-resolution video camera inside the car detects traffic signals, as well as pedestrians, bicyclists, and other moving obstacles. The cars also track their positions with a GPS and an inertial motion sensor.

Article

Google surprised the world last October, when the search engine company revealed it operated a fleet of seven autonomous cars that had driven a total of 140,000 miles. More stunning, perhaps, was that some of those cars had made it down the tight turns of San Francisco's Lombard St., crossed the Golden Gate Bridge, and circumnavigated Lake Tahoe—all without any human intervention.

Yet the search engine giant had hidden its self-driving cars in plain sight. The cars themselves were easy to identify by the spinning sensor on their roof. And while Google's six Priuses and one Audi TT all had someone in the driver's seat, their hands were clearly not on the wheel.

What was even more visible, however, was the technology. Many of the innovations that drove Google's autonomous vehicles have been publicly discussed since Defense Advanced Research Projects Agency's first challenge for driverless cars in 2004. In fact, a review of the DARPA challenges shows just how fast autonomous vehicles have advanced: No teams completed the 150-mile desert course that first year, but by 2007, six teams completed the more demanding 60-mile DARPA Urban Challenge, which required autonomous vehicles to traverse an abandoned military base, obeying traffic lights and stop signs, avoiding obstacles, yielding at intersections, and merging with traffic.

This customized Toyota Prius drove itself through the California countryside

Grahic Jump LocationThis customized Toyota Prius drove itself through the California countryside

Three components make Google's driverless cars go: sensors, software, and Google's mapping database. The sensors are configured much like the ones used by teams in the DARPA Urban Challenge, but most are neatly tucked away in the car's body rather than mounted, laboratory-style, on a roof rack.

The exception is the rotating sensor mounted on the roof. It is a Velodyne high-density LIDAR–light detection and ranging–that combines 64 pulsed lasers into a single unit. It was invented by Velodyne founder David Hall, a DARPA Challenge alumnus, and used by five of the six Urban Challenge finishers. The system rotates 10 times per second, capturing 1.3 million points to map the car's surroundings with centimeter-scale resolution in three dimensions. This lets it detect pavement up to 165 feet ahead or cars and trees within 400 feet.

Automotive radars, front and back, provide greater range at lower resolution. A high-resolution video camera inside the car detects traffic signals, as well as pedestrians, bicyclists, and other moving obstacles. The cars also track their positions with a GPS and an inertial motion sensor.

Blending information from different sensors requires very smart algorithms. Even then, AI is nothing more than educated guesswork. “If it's the right size and it moves, it's likely to be car,” said Michael Montemerlo, a Stanford software engineer working on the Google car project, in a 2007 interview. Yet a car that stops to back into a parking space may still cause confusion.

Google's AI even has multiple personalities, driving cautiously or aggressively at the switch of a switch. If several cars stop at an intersection, the cautious AI will wait its turn until it is really, really sure it can go.

The third key element of Google's driverless car, the company's mapping database, not only maps local roads, but also generates street-level pictures of them.

Before each test drive, a conventionally driven car maps the route and road conditions, said Sebastian Thrun, the director of Stanford's artificial intelligence lab and leader of Google's autonomous vehicle program. When the autonomous car travels the route, it refers to the database and updates it as it goes. Surprisingly, Google engineers discovered that the roads often changed between the mapping and test run.

Clearly, the technology is years away from implementation. Thrun and other Google engineers believe robotic cars could reduce accidents, lower energy consumption by more efficient driving, and improve the capacity of highways by letting cars safely drive closer to one another.

Montemerlo sees autonomous cars as opening the door to a robotic future. Right now, he said, increasingly computerized cars are simply “robots without the software to make them drive.” Montemerlo's Google team is intent on adding that software.

Copyright © 2011 by ASME
View article in PDF format.

References

Figures

Tables

Errata

Discussions

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging and repositioning the boxes below.

Related Journal Articles
Related eBook Content
Topic Collections

Sorry! You do not have access to this content. For assistance or to subscribe, please contact us:

  • TELEPHONE: 1-800-843-2763 (Toll-free in the USA)
  • EMAIL: asmedigitalcollection@asme.org
Sign In