November 6, 2007

DARPA Urban Challenge: Innovation Seedbed for 3D Laser Scanning

DARPA Urban Challenge - Image 3

Last Saturday the final round of the DARPA Urban Challenge took place at the former George Air Force Base in Victorville, CA, now used by the military to train for urban operations. On the high desert northeast of Los Angeles, 11 cars bristling with computing power, laser scanners, GPS receivers and other sensors navigated winding streets, merged into moving traffic, negotiated intersections, and interacted with manned vehicles and with one another – all operating autonomously. I was part of an audience of several hundred who watched Boss, a vehicle fielded by the Carnegie Mellon/GM Tartan Racing team, claim the $2 million first prize. Stanford Racing‘s Junior won the $1 million second prize, while Victor Tango‘s Odin from Virginia Tech took the $500,000 third prize. The winners were selected based on how quickly they navigated the course, and also how safely. 

This was the third in a series of competitions sponsored by DARPA to foster development of autonomous robotic ground vehicle technology. DARPA’s interest is in technology that will let the military remove troops from hazardous situations on the ground, much as unmanned aerial vehicles are already making possible in the skies. Our interest – autonomous vehicle guidance is a field where many of the sensing and processing technologies used in 3D laser scanning originated, and it continues to be a seedbed of innovation for the industry.

Continued

Progress of the DARPA Challenges

In DARPA’s first Grand Challenge, held in 2004 on a 142-mile desert course between Barstow, CA and Primm, NV, 15 vehicles attempted the course but none finished. DARPA Director Dr. Tony Tether – a hands-on presence throughout this year’s finals – likened that event to the Wright brothers’ first flight, “where their airplane didn’t fly very far but showed that flight was possible.” In the 2005 Grand Challenge, four vehicles successfully completed a 132-mile desert route in southern Nevada within a 10-hour time limit, and DARPA awarded a $2 million prize to Stanley, a vehicle from Stanford University.

In this year’s event, each vehicle was tasked with completing a complex 60-mile urban course with live traffic within six hours while obeying California traffic laws. Driving challenges included traffic circles, four-way intersections, blocked roads, parking, passing slower vehicles, and merging safely with traffic on two- and four-lane roads. From the time each vehicle left its starting chute, it was entirely under the control of its onboard mission computer; human intervention was allowed only for purposes of safety, such as when one vehicle nearly collided with a building.

The whole fleet of robotic vehicles was on the course at the same time. Thus they had to cope with two kinds of live traffic – not only some 50 human-driven cars (outfitted with NASCAR-level driver protection cages), but also one another. Just before launch, DARPA’s Dr. Tether emphasized that what we were about to witness – “robots interacting with one another” – would be “a first, and more unpredictable” than anything in the challenges thus far. In what I saw, many vehicles demonstrated impressively complex behavior – one of the most dramatic moments came when two cars collided, paused, then moved apart and both went on to complete the course.

Technical challenges

I asked first-place Tartan Racing’s Chris Urmson, Director of Technology for the Urban Challenge at Carnegie Mellon’s Robotics Institute, what he considered the greatest technical challenge. He began by listing three technical requirements of this year’s event – “driving down the road, parking, and reasoning at intersections.” Unlike past DARPA challenges, where each vehicle had only to navigate terrain, the addition of the second and third tasks “made this year’s challenge enormously more complex than ever before.” Calling it a “great achievement that 11 teams made it to the finals, and six vehicles completed the event,” Urmson added that an incalculable value of these technologies is their promise eventually to slash civilian traffic fatalities, besides their military uses. [We’re honored to welcome Urmson as keynote presenter at SPAR 2008, March 3-5, Houston, TX.]

Sebastian Thrun, team leader of second-place Stanford Racing and director of Stanford University’s AI Lab, told me that a key technical challenge was balancing the competing requirements of operating rapidly but safely – how far could safety constraints be relaxed to gain some speed? In a post-race panel discussion, Thrun also raised the notion of a “driving Turing test” – where an observer, viewing airborne video of a vehicle’s behavior, would try to decide whether the driver was a human or a robot. “I think that in a couple of years,” he predicted, “you won’t be able to tell.” DARPA’s Dr. Tether went further, declaring that in scrutinizing this year’s helicopter-captured videos as part of the judging process, he often thought the robotic vehicles were moving just as they would if driven by a person.

Continued

Innovation hotbed

Innovation abounded in the Challenge teams’ pits. Ibeo Automobile Sensor GmbH, the Hamburg-based automotive subsidiary of SICK AG, showed me the LUX, a prototype laser scanner it developed for Team LUX, its own entry in the Urban Challenge. Team LUX was led by intelligent vehicle specialist Dr. Richard Bishop, Bishop Consulting, Granite, MD, and included current and former employees from Ibeo and SICK. The LUX is a time-of-flight laser scanner packaged in a sealed unit measuring 128mm x 85mm x 83mm. FOV is 150deg horizontal by 32deg vertical. Reported range is 0.3m-200m, and reported accuracy is 4cm. “Our aim,” technical team leader Holger Salow told me, “is to have this sensor in all cars!”

We think Ibeo’s pricing plans are a bellwether of what will happen when laser scanning technology achieves penetration in volume markets. Prototype versions are available today for €40,000, sales and service director Mario Brumm says. In October 2008, production versions are planned to be available for €10,000 list price. However, “if we get a volume contract from a GM or a Toyota,” Brumm predicts, “unit price will be €200 in the 2010-2012 time frame.” Uptake of laser scanning technology in automotive and other mass markets promises to be the breakthrough that in time will let prices of terrestrial laser scanning technology come down dramatically – as well as driving new levels of product robustness and compactness.

Another new product that drew considerable attention was a laser scanner from Velodyne Acoustics, Inc., Morgan Hill, CA. The system was used by 7 of the 11 finalist teams, according to Velodyne president Bruce Hall, and 12 of the 35 semifinalists. The Velodyne system, a rotating cylinder mounted on the vehicle roof, uses 64 lasers to capture 1 million points/second at position accuracy of +/- 1 inch and better than 1/10deg rotational accuracy, Hall says. FOV is 360deg horizontal by 2deg above grade to 24deg below grade vertical. Unit price is $75,000, and “everyone who’s bought one paid that price,” Hall smiles. Sales to date are in the tens of units, he reports.

Velodyne, a maker of audio speaker systems, has been involved with laser scanning since the 2005 DARPA Grand Challenge. President Bruce Hall says this was driven by his brother, Dave Hall, developer of Velodyne’s acoustic woofer technology and “a prolific inventor” who became interested in robotic vision and autonomous vehicle navigation in the late 1990s. From its experience in the 2004 DARPA challenge, “we became unhappy with stereovision approaches, and went into LIDAR.” Hall reports automotive OEMs have expressed interest in Velodyne’s scanning technology, including inquiries about integrating the sensor unit into the industrial design of the vehicle.

Ibeo and SICK lasers were used in many of the 2007 finalists. I saw Riegl laser scanners on numerous vehicles as well.

Applanix navigation systems were used by 8 of the 11 finalists, according to Louis Nastro, Applanix director of land products. In addition, he says, two other finalist teams that developed their own navigation systems purchased Applanix units as backups.

Dr. Norman Whitaker, Urban Challenge Program Manager, noted the great spinoff value of the technologies pioneered here, as well as the contest’s value in attracting new students to these fields. “The community-building taking place is key,” he told us in a predawn press briefing. “Lots of our university teams – we can hardly get them to sleep at night. And university deans contact us to tell us, ‘This is best thing you’ve ever done,'” reporting a “huge influx” of applications to their engineering, computer science and other programs after their teams participated in past Challenge events.

Team profiles

Tartan Racing, Boss, Pittsburgh, PA – Carnegie Mellon University’s Robotics Institute came together with General Motors to form Tartan Racing. Team members were employees of the Robotics Institute, and other departments within Carnegie Mellon University and General Motors. Sponsors: Carnegie Mellon, Caterpillar, Continental, GM, Google, IBEO, Intel, McCabe Software, MobilEye, NetApp, Tele Atlas, Vector CANTech and Viewpoint.

Stanford Racing, Junior, Stanford, CA – The team was drawn from faculty and students at Stanford University’s School of Engineering and sponsoring corporate partners. Sponsors: Android, Applanix, Coverity Inc., Google, Honeywell, Intel, Mohr Davidow Ventures, NXP, Red Bull, Tyzx, Inc. and Volkswagen of America Electronics Research Lab.

Victor Tango, Odin, Blacksburg, VA – Victor Tango, from Virginia Tech, consisted of undergraduate and graduate students and faculty, paired with a Virginia Tech autonomous systems spinoff company, Torc Technologies. Sponsors: Black Box, Caterpillar, Ford, GM, Goodyear, Honeywell, IBEO, Ingersoll Rand, Lockheed Martin, Michelin, National Instruments, NovAtel, OmniSTAR, QCI, SICK, Tripp-Lite and Ultramotion.

Want more stories like this? Subscribe today!



Read Next

Related Articles

Comments

Join the Discussion