February 19, 2007 The 2005 DARPA Grand Challenge marked a milestone in artificial intelligence when five autonomous vehicles finished the course and Stanford Racing Team’s Stanley went down in history as the winner of the first race for autonomous vehicles. Centuries from now, the win will be equally significant as winning the first auto race from Paris to Rouen in 1894. Indeed, a century from now, there’s every chance that cars will all be autonomous, as computers make less mistakes than human beings. The robots in the 2007 Urban Challenge, however, will need all of Stanley’s capabilities plus a whole lot more as this time they need to handle real traffic. “In the last Grand Challenge, it didn’t really matter whether an obstacle was a rock or a bush because either way you’d just drive around it,” says Stanford Team Leader Sebastian Thrun. “The current challenge is to move from just sensing the environment to understanding the environment.” Thrun is the Director of the Stanford Artificial Intelligence Lab and Associate Professor of Computer Science at Stanford University. When the bookmakers frame the odds for the Urban Challenge, Thrun’s charge will be favourite. On Saturday, Thrun introduced Stanford Racing Team’s new challenger to the world. Junior is a new generation of autonomous vehicle built to accomplish missions in a simulated city environment, which includes the traffic of the other robots and traffic laws. This means that on race day, November 3, Junior not only will have to avoid collisions, but he will have to master concepts that befuddle many humans, such as right of way. Junior began life as a 2006 Volkswagen Passat wagon.
“This has a component of prediction,” says Mike Montemerlo, a senior research engineer in the Stanford Artificial Intelligence Lab (SAIL). “There are other intelligent robot drivers out in the world. They are all making decisions. Predicting what they are going to do in the future is a hard problem that is important to driving. Is it my turn at the intersection? Do I have time to get across the intersection before somebody hits me?”
Racing team leaders Thrun and Montemerlo discussed Junior for the first time Feb. 17 at the annual conference of the American Association for the Advancement of Science in San Francisco. Thrun joined fellow roboticists in a panel discussion “Robots—Our Future’s Sustainable Partner” at 8 a.m. He spoke about autonomous guidance systems and machine vision. Afterwards, he and Montemerlo participated in a press conference at noon. Introducing Junior
Junior is a 2006 Passat wagon whose steering, throttle and brakes have all been modified by engineers at the Volkswagen of America Electronics Research Lab in Palo Alto, Calif., to be completely computer-controllable. The engineers also have created custom mountings for a bevy of sophisticated sensors.
An important difference between Junior and Stanley is that Junior must be aware of fast- moving objects all around it, while Stanley only had to grapple with still objects in front of it. Junior’s sensors are therefore much more sophisticated, Thrun says. They include a range-finding laser array that spins to provide a 360-degree, three-dimensional view of the surrounding environment in near real-time. The laser array is accompanied by a device with six video cameras that “see” all around the car. Junior also uses bumper- mounted lasers, radar, Global Positioning System receivers and inertial navigation hardware to collect data about where it is and what is around.
Because Junior collects much more data than Stanley did, its computational hardware must be commensurately more powerful, says Montemerlo. Using Intel Core 2 Duo processors—each chip includes multiple processing units—Junior’s “brain” is about four times more powerful than Stanley’s. But what makes Junior truly autonomous will be its software, which is the focus of about a dozen students, faculty and researchers at the SAIL. Modules for tasks such as perception, mapping and planning give Junior the machine-learning ability to improve its driving and to convert raw sensor data into a cohesive understanding of its situation.
New software development began last fall. Montemerlo has been testing some of the team’s software modules in simulated traffic situations since the beginning of the year.
The team expects to move into full-time testing and iterative improvement by the end of March.
Junior’s name is not only an implicit homage to its predecessor, but also to Stanford University’s namesake, Leland Stanford, Jr. Carrying this sense of history, Junior will set out to make technology history of its own and pave the way to a future where autonomous cars can make driving safer, more accessible and more efficient. Self-driving cars could give drivers newfound free time.
“You could claim that moving from pixelated perception, where the robot looks at sensor data to understanding and predicting the environment, is a Holy Grail of artificial intelligence,” says Thrun. Junior’s name is not only an implicit homage to its predecessor, but also to Stanford University’s namesake, Leland Stanford, Jr. Carrying this sense of history, Junior will set out to make technology history of its own and pave the way to a future where autonomous cars can make driving safer, more accessible and more efficient. Self-driving cars could give drivers newfound free time.
“You could claim that moving from pixelated perception, where the robot looks at sensor data to understanding and predicting the environment, is a Holy Grail of artificial intelligence,” says Thrun.
The racing team, based in the School of Engineering, is supported by returning industry team members Intel, MDV-Mohr Davidow Ventures, Red Bull and Volkswagen of America and joined this year by new supporters Applanix, Google and NXP Semiconductors. DARPA also has provided $1 million of funding.
Make and model: 2006 Volkswagen Passat wagon Engine: 4-cylinder turbo diesel injection Transmission: Six-speed direct shift gearbox Engine cubic capacity: 1968cc Fuel Consumption: City: 25.5 mpg (9.2l/100km) Highway: 42.7 mpg (5.5l/100km) Combined: 34.6 mpg (6.8l/100km) Power: 140 hp (103kW) at 4000rpm Torque: 236 lb ft (320Nm) at 1800-2500 rpm Top speed: 126miles/h (203km/h) Acceleration 0-100km/h: 10.1sec Power is provided by the engine through a high-current prototype alternator and a battery-backed, electronically-controlled power system. The senses
Position: Junior's position and orientation are determined by a cutting-edge Applanix POS LV 420 system that is optimized for adverse GPS environments. The system provides real time integration of multiple dual-frequency GPS receivers, a high- performance inertial measurement unit (IMU), wheel odometry, and Omnistar's satellite- based Virtual Base Station (VBS) service. Real time accuracy exceeds 35cm and 1/50th of a degree. Sight: is provided by several state-of-the-art sensors. A Velodyne HD Lidar looks in every direction at once. It combines 64 individual lasers into millions of 3D points per second at up to 50m range. An Ibeo ALASCA XT Lidar handles long ranges, with four scanning planes reaching as far as 200m. A Point Grey Ladybug 2 provides six video cameras that produce near-high-definition video in every direction. SICK Lidar scanners (which Stanley used in 2005) are used for precision navigation at low speeds.
Hardware: Provided by rackmount servers equipped with Intel's latest Dual and Quad Core processors. Data is processed from instruments as frequently as 200 times a second. Software: Integrated, custom-coded modules include a planner (making decisions, choosing routes), a mapper (transforming sensor readings into environment understanding), a localizer (refining GPS position by visual observations), and a controller (actuating the planner decisions on the car).