The Stanford Racing Team’s autonomous car, Junior, passed a complex driving test Thursday, making it one of the few robots in the world able to deal with the complexities of city traffic. Well, maybe small town traffic.
“It drives like my grandma,” exclaimed one bystander, as Junior cautiously pulled up to an intersection, turned on its blinker, waited ten seconds, and then pulled cautiously and jerkily around the curve.
Pathetic for a human — but pretty damn impressive for a completely self-contained, autonomous robot.
Junior is a highly modified 2006 Volkswagen Passat with an array of spinning lasers mounted on its roof and fenders and a pair of servers in its trunk. It is the successor to Stanford’s earlier robot, Stanley, which won the DARPA Grand Challenge in 2005 by successfully navigating a 132-mile desert course.
Junior is built for a more difficult mission: DARPA’s Urban Challenge, which will put a fleet of competing robot vehicles onto a mock urban streetscape, where they’ll be expected to navigate around obstacles, recognize and obey traffic signs, negotiate intersections, and avoid collisions with other cars, including their competition.
“The Urban Challenge subsumes [last year’s] desert challenge,” said Mike Montemerlo, one of the Stanford team leaders and a professor of computer science at the university. That’s because Junior has to do everything Stanley did — follow a course and avoid obstacles — plus it also has to detect other vehicles and anticipate their behavior in traffic.
The team is betting on an array of technologies to take on this challenge. The Passat was chosen because its control systems are almost entirely electronic, making it a relatively simple matter to add a “drive-by-wire” system — an electronic system that allows a computer to control the car. The Stanford engineers just patch a Category 5 cable into the system, and their computer drives the car by passing UDP data packets to it.
The computers powering Junior are two rack-mounted Intel servers, one with a four-core Intel Core2 Quad chip, and another one with a dual-core Intel Core2 Duo.
The sensors include a roof-mounted laser rangefinding system that spins 15 times per second to build up a 360-degree view of its surroundings, out to a distance of about 65 meters. Additional lasers on the car’s corners give Junior detailed data about nearer objects, such as curbs. Positioning data comes from a module supplied by Applanix, which uses three accelerometers and three gyroscopes, plus data from wheel sensors, to augment the car’s GPS receiver.
All of the data is fed into the servers, where an array of nine separate software modules written by the Stanford team parses it, computes Junior’s course, and passes control signals on to the Passat. It’s that software that makes the difference between success and failure.
“We really think this is a software competition,” said team member David Stavens. “There are a lot of ways you can lose a competition — a popped tire or what have you — but to win it takes reliable artificial intelligence.”
Before teams can enter the qualifying trials this fall, they first must pass an on-site inspection by a team of DARPA representatives, and that’s what was happening Thursday in a parking lot just outside the Shoreline Amphitheater in Mountain View, California. Fifty-three teams have qualified for DARPA site visits.
Members of the Stanford team (blue shirts) confer with DARPA officials (white shirts) prior to the start of Junior’s driving test.
During the course of the morning, the Stanford team members put Junior through its paces before the watchful eyes of DARPA. First they tested Junior’s ability to stop on demand, by remote control. Then the car navigated itself through a simple course marked out with yellow lines and orange cones, stopping at intersections indicated by white bars on the pavement. The next test involved driving through the course and passing around stopped cars. And the fourth test required Junior to approach intersections, wait while another car drove through it, and then drive through the intersection itself.
“Negotiating an intersection is a complicated thing,” said Anya Petrovskaya, a PhD candidate in computer science at Stanford, and the team member who wrote the software responsible for calculating other vehicles’ speed and direction. “People do all kinds of things like making eye contact and waving their hands, and Junior doesn’t have any of that information.”
Junior is an exceedingly conservative driver, acknowledged Montemerlo. Partly because of that conservatism, Junior got stuck during the third trial, where it stopped dead while passing a parked car, unable to go forward because it had insufficient clearance between the other car and the edge of its “road.”
All was not lost for the Stanford team, however. It turned out that they’d placed the cones too close together, making the roadway narrower than DARPA had specified. When the cones were placed at regulation distance and Junior’s minimum clearance distance adjusted, the robot car sailed through the trial.
The technology is not yet bulletproof. “I wouldn’t want to get out on the course with a bunch of robots — at least not yet,” said Montemerlo. Junior is programmed to handle traffic signs and other cars, not bikes, pedestrians, construction equipment, and other hazards of real-world traffic. “Being able to handle the chaos of really tough urban situations is a ways off, maybe 20 to 25 years,” Montemerlo said.
The technology being developed by Stanford and its corporate partners (including Volkswagen, Intel, Applanix, and NXP, among others) may lead to more intelligent driver-assistance systems in the coming years, and maybe, eventually, to completely autonomous passenger cars. DARPA is also interested in the possibilities of using unmanned autonomous ground vehicles in combat.
Team leader Sebastian Thrun envisions a day when humans are no longer required to drive their cars. “First of all, cars are unsafe. We kill something like 42,000 people per year, and most of those deaths are due to human error. Second, cars are inefficient. They require a lot of time and attention to drive… I think that autonomous cars will really change society.”
But before they can change society, the Stanford Racing Team have to get ready for this fall’s qualifying trials, assuming they pass the DARPA inspection (the results won’t be announced until August).
“We’ll probably do a lot of testing over the summer,” said Petrovskaya. “But we already have a good start.”
A view inside Junior’s trunk shows the robot’s brain: two rackmounted servers.
All photos by Walden Kirsch, Intel.
Zakeh Blog
Researchers at HP Labs have built the first working prototypes of an important new electronic component that may lead to instant-on PCs as well as analog computers that process information the way the human brain does.
Internet phone company Skype has 
In Las Vegas Sprint has announced a new mobile called the Instinct made by Samsung. It weighs in and is about the same size as the iPhone, and seems to use similar interface ideas. One of the most touted features of the Instinct over the iPhone is the ability to run on EVDO Rev. A, which can achieve much faster speeds than AT&Ts EDGE network. Instinct will come with a 2GB microSD card and it can be expanded up to 8GB. It also features a 2 megapixel camera, GPS navigation (using Telenav), Visual Voicemail (using the touch screen to scan through and browse voicemail messages), and two 1000 mAmp batteries giving up to 5.75 hours of talk time each, as well as some other accessories.
Fujitsu Limited today announced its new MHZ2 BJ series of 2.5″ hard disk drives that feature a rotational speed of 7200 RPM and capacities up to 320 GB Sales of the new series will begin at the end of June 2008. The new series is designed for use in high-performance PCs, such as high-end notebook computers and compact desktops.
The end of the high-definition disc war seems in sight, say industry observers, with major studio Warner Bros. jumping to the Blu-ray Disc camp and leaving the competing HD DVD format with a dwindling base of Hollywood support.
