Global View  November 2005

November 1, 2005 GPS World

Robot on the Run — Stanford’s Stanley Takes DARPA Grand Challenge Prize

Stanley, the robotic Volkswagen Touareg from Stanford University, won the 2005 Grand Challenge mounted by the Defense Advanced Research Project Agency (DARPA). With no one at the wheel, Stanley covered the 131-mile Mojave Desert course in 6 hours, 53 minutes, 8 seconds, at an average speed of 19.1 miles per hour — good enough for $2 million. Stanley incorporates measurements from a NovAtel Propak LB GPS receiver with OmniStar HP differential corrections, an Inertial Sciences Inc. ISIS inertial measurement unit, a Novatel Beeline GPS compass, and wheel odometry from the Touareg’s CAN bus.

In motion, Stanley perceives the environment through five laser rangefinders, a monocular video camera, and radar for long-range sight. The sensors acquire data at rates of 10 to 100 Hz. Map and pose information are incorporated at 10 Hz. Processing takes place on six Pentium M motherboards in a rugged rack mount unit.

Despite this impressive list of hardware, “We treated this as a software problem,” said Mike Montemerlo of the Stanford racing team. “The first year [of the Challenge], people did a lot of hardware application. But I basically saw it, the critical part, taking place in software.”

Montemerlo wrote an unscented Kalman Filter (UKF) to integrate the measurements and generate a 6-degrees-of-freedom (DOF) pose of the vehicle. The output of the UKF and the laser data serve as inputs to Stanley’s “mapper,” which outputs an obstacle map to “the planner,” which plans a safe trajectory and outputs it to “the controller,” which sends throttle, brake, and steering commands to the Touareg, enabling it to avoid obstacle collisions obstacles in real time while advancing along the route.

“The UKF is a version of the extended Kalman filter (EKF) that handles nonlinear systems more accurately than the EKF,” said Montemerlo. “It’s also easier to implement correctly. When you have these multidimensional nonlinear systems, computing derivatives can be long and complicated. In my own experience, I frequently got it wrong. The unscented filter computes derivatives numerically. You put in your motion model and your measurement model of all your sensors, and it takes care of computing the derivative automatically.”

Overtaken at the Pass. Stanley did not grab the lead from his nearest competitor until most difficult obstacle, Beer Bottle Pass, seven miles from the finish line. Extremely steep, with a mountain on one side of the road and a 100-foot drop on the other cliff, Beer Bottle Pass represented “a catastrophic failure mode,” recalled Montemerlo. “From a GPS perspective, that was probably the worst environment because you don’t have a complete view of the sky, and very little tolerance, maybe 10 feet, in your zone.”

Advancing at 10 miles per hour, Stanley grazed the edge of the road a couple of times, causing onlookers to hold their breath. But he completed the climb and roared down to the finish in Primm, Nevada, where team members, journalists, and officials had followed the race by video feed and 3D mapping projectors, in huge tents erected in a parking lot surrounded by casinos and hotels.

Two vehicles outfitted and trained by Carnegie-Mellon University’s Red Team finished just minutes behind, when all official pauses were reconciled and computed. The veteran Sandstorm came in at 7:04:50 and race rookie H1ghlander at 7:14:00. Both vehicles used Applanix POS LV, Trimble Zephyr antennas and AG 252 GPS receivers, and OmniStar correction data. The Gray Team’s KAT-5 (7:30:16) used an RT3000 inertial/GPS unit from Oxford Technical Solutions, which incorporated a NovAtel receiver and OmniStar corrections. Oshkosh Truck’s 16-ton TerraMax, with Trimble AgDGPS and OmniStar, finished the next morning.

CLARIFICATION

In an August story, “Melee in the Mojave,” we identified another GPS receiver and differential service for Stanley. Sometime during the summer, the Stanford team made a change to those listed here.

Nobel Prize Has GPS Implications

The Nobel Institute awarded its 2005 Prize in Physics to John L. Hall, a senior research associate at the Joint Institute for Laboratory Astrophysics (JILA) at the University of Colorado and the National Institute of Standards and Technology (NIST), and to Theodor W. Hänsch of the Max Planck Institute for Quantum Optics and Roy J. Glauber of Harvard University. Hall and Hänsch’s work has the potential to develop extremely accurate clocks and, ultimately, to improve GPS technology.

Hall and Hänsch received half the physics Nobel for development of laser-based precision spectroscopy, including the optical frequency-comb technique. Frequency comb describes the spectrum of the output of a mode-locked laser that emits a stream of pulses and permits the measurement of optical frequencies, which otherwise is very difficult. Basically, any optical frequency can be converted to a countable radio frequency.

Current practical atomic clocks (rubidium, cesium, and hydrogen maser) are based on an atomic energy transition at a radio frequency that can be measured easily with electronic devices. A better clock could be fashioned by using an atomic or molecular transition at a much higher frequency in the optical part of the electromagnetic spectrum. The higher the frequency, the higher the clock stability. By locking the frequency comb to an optical transition, an optical atomic clock will have an output at a useful radio frequency.

Performance of optical clocks already approaches that of the best cesium standards in the world and likely will soon surpass them. A Japanese group has invented an optical atomic clock based on strontium atoms with the potential to keep time to one part in 1018. If practical optical atomic clocks can be constructed, they could be used in GPS or other GNSS satellites and at ground monitoring stations to improve navigation accuracy.

L1C Project Collects Signal Input

The L1C Project Team is now scheduling interviews with stakeholders and technical experts on proposed designs for the L1C civil signal (see “Position,” page 52 in this issue, for details). Upon completion of the interview stage, the project team will incorporate findings into a signal definition document, in the form of a draft L1C Interface Specification (IS), by March 1, 2006. This will allow timely incorporation of L1C into the GPS III contractual specifications at the GPS Joint Program Office (JPO).

The completed L1C IS draft will enter the Public Interface Control Working Group (ICWG) process at the GPS JPO. The ICWG process allows input by international participants and maintains open information for all users, as with existing documents, IS-GPS-200 (for L1 and L2) and IS-GPS-705 (for L5). Anyone interested in participating in L1C signal design after submittal of the L1C IS in March 2006 should join the ICWG (see http://gps.losangeles.af.mil/engineering/icwg).

The L1C Project has been supported for the past two years by the Interagency GPS Executive Board (IGEB) through its GPS Stewardship funding, and through in-kind support from the U. S. Air Force GPS JPO and the Federal Aviation Administration.

General Dynamics Studies Ground Control

General Dynamics C4 Systems has received a contract from the U.S. Air Force to study hardware, software, and technology options to deliver an advanced GPS satellite control segment architecture. The contract is valued at $1.4 million.

This GPS Operation Control Segment (OCX) program, when fully developed, will operate the GPS III satellite constellation, legacy GPS satellites and other military satellites as part of an integrated space/ground network.

The six-month study will investigate new plug-and-play methods for controlling GPS satellites in space and support creation of the Air Force’s requirements for the GPS OCX program. After the six-month study, the Air Force is expected to award a contract for the development of the next-generation GPS ground segment, with an estimated value of $500 million to $1 billion over ten years.

Bookmark it: digg propeller del.icio.us technorati yahoo facebook

Add Comment