Prod_GI_2_2006 28-02-2006 10:16 Pagina 42

Special TheThe DARPADARPA Grand Grand ChallengeChallenge andand itsits LegacyLegacy Moving Technology Forward

world when prototype “auto-control” (almost robotic) cars traveled bumper-to-bumper in platoons along special highway lanes while For inventors and future capitalists alike, some of the greatest moments in the twenti- drivers read newspapers. Road development attentions then shifted focus towards infras- eth century began as irresistible contests. From Raymond Orteig’s challenge to fly tructure designs, driver/roadway information management systems, and command and between New York and Paris non-stop to the now famous Ansari X-Prize for the first control centers. It would take more than ten years of private manned spacecraft, the end reward is greater for progress than for those who advancement before suitable technologies would be available for the spotlight to return first succeed. to vehicle automation. It would also take abolishing selective availability on GPS sig- By Anthony Melihen and Louis Nastro nals and losses in the battlefield plus the acknowledgement of critical human driving weaknesses before this technology would be Ultimate Challenge of engineers. The course could not be longer actively pursued again with a practical goal In October 2005, history added the DARPA than 175 miles and would not be revealed to in mind. Grand Challenge to the list – possibly the participants until just prior to race time. The ultimate technological challenge to date, like- first team to finish in less than 10 hours Before Sunrise ly the one that will have the greatest impact wins. October 6th, 2005, the Mojave Desert near on our daily lives. It is also certainly the Originally conceived in 2001, the first DARPA Primm, . Perhaps the most important safest to the contenders, and consequently Grand Challenge was held on March 13, morning commute ever taken was about to such a remarkable beginning will be 2004. The furthest distance traveled that day begin. Before the sun would rise, competi- applauded by only a few outside of the land was only 7.6 miles and the $1 million dollar tors would begin a demonstration to the transportation industry. prize sat unclaimed. In 2005, the purse dou- world of how far transportation has The Defense Advanced Research Projects bled and word spread internationally. 195 progressed and what might be in the years Agency (DARPA) Grand Challenge is a compe- teams, twice as many as in 2004, would ahead. tition to create and successfully race a enter. Only 43 would be chosen to qualify, Each of the 23 vehicles would have to prove driverless vehicle across the bitter desert 23 would race as finalists. themselves from the start. A form of robotic environment with natural and man-made vehicle cross training, racers would face obstacles in place. These are designed to Infrastructure Designs identical yet varying conditions designed to interrupt or distort GPS signals, break wheel A decade earlier, the concept of robotic vehi- test all major data processing and response struts, and humble even the most boastful cle assistance was demonstrated to the capabilities. Navigation and sensor impedi-

42 March 2006 Prod_GI_2_2006 28-02-2006 10:16 Pagina 43

Special gege andand itsits LegacyLegacy

Sandstorm successfully negotiates the exact middle path around a hillside switchback while the chase vehicle follows.

ments – manmade and natural – were in repeatedly tested their robotic trucks in was an essential part of the CMU race strate- place throughout the course. There were also extremely harsh conditions with the expec- gy, it compared optimal conclusions with a the distance requirements, the race time tation of racing on a tougher off-road dataset library. Accordingly the processor limit, the desert terrain, and extreme varia- course. Some of CMU’s advanced research came up with an action conclusion that tion in daily temperature to deal with. Even in path-finding scanning systems may have obeyed the race parameters. the predawn start time was calculated to seemed over engineered given the down- frustrate weaker optical sensors due to low graded course changes, and therefore Good GPS Essential angle sunlight. As a final twist, teams were would not be as adequately challenged as Forward scanners like , laser, or a vari- given the promised waypoints for the CMU had hoped. This year’s challenge ety of choices, were indispensable to all announced 131.6 mile (211.8 km) course only seemed custom designed to test of how designs on the field. A good GPS was equal- two hours before their designated start time. navigation systems could follow a well- ly essential. However, hopeful entries that All combined, any vehicle that successfully defined path. Race strategies were changed opted for a lighter design with vehicle posi- completed this challenge could claim the day overnight. tion awareness responsibilities assigned to a as a remarkable victory. GPS unit alone would be disappointed. An Sandstorm and H1ghlander advanced GPS system operating closely with Waypoints The two CMU vehicles consisted of the beefy forward scanners but without added posi- Once the course had been driven by a pro- HUMMER cousins, both with bold high- tional support is subject to far too many fessional desert race driver, DARPA set the mounted sensor arrays and bright red paint. variables to be considered reliable for robot- maximum design speed for completion at 6 Their names: “Sandstorm”, a 1986 model ic vehicles. Even given relatively fast GPS hours and 18 minutes, an average course 998 HMMWV and the veteran DARPA dis- refresh rates of approximately 1 Hz, robotic speed of 20.9 mph (33.6 km/h or 9.3 m/s). tance record holder, and “H1ghlander”, a travel at higher speeds remains unsafe Over 2,930 waypoints were provided; an 1999 H1 HUMMER and the pole position because in that short moment, a leisurely average of 1 waypoint for every 72 meters holder. paced vehicle can cover a significant dis- with points varying in separation from a Risk adverse CMU engineers adopted a sys- tance without guidance. few meters to 500 meters apart. A chase tem approach with redundancy and back-ups Anticipating an upcoming turn after a suc- truck would follow each entry and a human for both Red Team vehicles. Scanning lasers, cessful GPS reading and before the next sud- spotter would stop any vehicle experiencing cameras, and radar technology were relied denly places this vehicle a dozen meters difficulties. Almost an afterthought, the upon to produce a composite model of the ahead or behind the sought waypoint in only course had been groomed by DARPA terrain ahead. Applanix POS LV technology one second. At that critical moment, a vehi- graders, creating the impression that the with inertial/GPS and distance measurement cle’s cooperating optic sensors may confuse conditions this year were considerably easi- hardware generated position and orientation a precipice as a pathway because it believes er than expected. data throughout the course. A processor (according to map data) it is still positioned Some teams like Carnegie Mellon’s Red simultaneously received and assessed data in the middle of a flat featureless plateau Team, with two vehicles entered, had from all systems. Since mission preplanning rather than on the edge.

Latest News? Visit www.geoinformatics.com March 2006 43 Prod_GI_2_2006 28-02-2006 10:16 Pagina 44

Special

through a triad of accelerometers and gyro- scopes, supplying data in six axes. Even if only one GPS signal is available following initialization, the Applanix navigation system will maintain continuous sub-meter position accuracy that can successfully guide a vehi- cle. Sub decimeter results with Real Time Kinematic (RTK) GPS are also possible. All vehicle movements (like tilt, turn, lean, and acceleration) are accounted for accurately and used for making fast navigational adjustments relative to the last GPS reading. This is ideal for keeping a vehicle on track in between GPS cycles or during extended blockages such as when traveling through a long tunnel; a challenge-ending obstacle that would appear twice on the DARPA course. Just like closing one’s eyes and trying to con- fidently walk a lengthy, perfectly straight path, an IMU’s position performance deterio- rates over time if unsupported. A specialized Urban Road Inertial Measurement Unit Distance Measuring Instrument (DMI) pro- Rugged, hilly, and rocky desert terrain further The CMU engineers adopted the Applanix vides accurate travel distance data when a complicate matters as lost GPS signals POS LV system. The workhorse component GPS signal is unavailable to do the job. This (blocked line of sight) and false GPS signal of the new system is its inertial measure- precaution reduces system navigation error echoes (reflected off canyon walls) produce ment unit (IMU). Combined into a single over prolonged GPS outages. position readings that are wrong or unusable, solution with a proven GPS and a sophisti- ending all hopes of a win, or possibly cated distance measurement instrument Raw GPS Observables retrieval. In an urban road environment, these (DMI), highly accurate position data could be Adding time-matched data for precise distance, GPS challenges become amplified while the produced, also during extended GPS signal speed, direction, and other navigational inputs room for error drops from feet to inches and outages. But most important was the critical improved the overall movement accuracy for from seconds to milliseconds. orientation data derived from the POS. both Red Team vehicles. Raw GPS obser- Teams that could invest in alternate position Collecting vehicle’s dynamics data in real vables are incorporated into the positioning and orientation technology to cooperate with time is essential not only for performance, solution based on their quality. This is why their vehicle’s GPS discovered that they were but for proper orientation of the sensors rel- it is often referred to as a tightly coupled holding a trump card. It also seemed clear ative to the vehicle’s motion. inertial GPS processing system. Each signal that, when combined, such systems could be The concept behind the POS LV system is is weighted and compared to the IMU and accurate and reliable enough to be used inde- relatively simple even if the hardware is not. DMI data. All combined, the system essen- pendently when externally referenced naviga- Modern advanced inertial measurement sys- tially checks itself continuously and identifies tion systems failed, even for short periods. tems can record slight changes in motion unreasonable readings through rapid

Applanix POS LV System, featuring GPS, IMU, POS Computer System, and DMI.

44 March 2006 Prod_GI_2_2006 28-02-2006 10:16 Pagina 45

Special

design of our roadways; a role that today may seem overwhelming. Next generation road networks will be data gluttons and therefore surface information will need to be frequently surveyed and conveniently assem- bled for wide access and dissemination. Rapid repeated coverage and vast storage support will need to be coupled to seamless communications networks in ways not even conceivable today. Road data may be collect- ed by robotic vehicles themselves, contribut- ing to other sources including aerial drones performing daily auto-orthorectified digital mapping of traffic corridors and perpetually

Awaiting the start in the sunrise. active robotic lane calibration vehicles. Government, military, and commercial trans- comparisons from all sensors. Critical Technologies port industries may drive early robotic vehi- If adding inertial and distance measurement In one contest, we have leapfrogged about a cle technology forward primarily due to eco- units to the GPS made the difference decade of relative stagnation. Within two nomic reasons. These sectors, however, will between success and failure, then the proof years, 23 test-ready vehicle designs and 5 also need to support its continued growth of the pudding is in the dust. Both Red successful prototype concepts were produced, through data provision, traditional surveying, Team vehicles finished, back-to-back, with varying from hybrid SUVs to a full-sized 16 research and development, the enforcement times of 7:04:50 and 7:17:00. The winner ton tactical cargo hauler. Critical technologies of standards with regulation, and expertise was the Stanford Racing Team who posted a such as inertial navigation systems with GPS as adoption grows. time of 6:53:58. Gray Team finished fourth technology prove how close we are to resolv- using a 2005 Ford Escape SUV hybrid ing major development hurdles, including Incredible Leap design. Team TerraMax’s Oshkosh Truck pinpoint vehicle navigation issues. Now, a lot The DARPA Grand challenge has done more earned fifth place credit, despite finishing of technological backfill is needed before we than demonstrated how a prize reward can after the 10 hour time limit had passed. Of learn if such development will even be become a tool for moving technology for- 23 vehicles qualified to run on race day, five accepted. ward. It has also placed the future of auto- completed the course; four finishing within Fewer accidents, safer roads, and better use mated robotic vehicles back in the spotlight. the time limit. And all five finalists were of time and lane space are benefits promised This event has moved critical development equipped with an inertial navigation unit. by futurists. Robotic cars can quintuple the forward by one incredible leap. After years of utility value of each highway lane by traveling seeing our nation’s roadways and vehicles Leading Edge Thinking closer together, each vehicle communicating simply as they are, we have struggled to a Who was the biggest winner? DARPA? For a with each other and with the roadway man- point where we can produce technology that comparatively small investment, DARPA agement system to maintain speed and safe- will change how we think about transporta- would be handed a wealth of innovative ty. Aerodynamic “drafting” flow from following tion and navigation. designs and unpublished volumes of leading bumper-to-near-bumper will cut gas consump- The preview has ended with five successful edge thinking – far more data and in less tion as well. Pressing our imaginations a bit challengers; we are ready to imagine what time than were they to develop a prototype harder, it may be conceivable that one day might follow. This vision of the future sud- internally or through a private tender. But impediments such as traffic lights and stop denly seems infinitely more reasonable now any investment might seem pointless with- signs may become unnecessary. and it may arrive sooner than we dreamed. out success; something organizers anxiously But before we get excited about 24 hour If we accept this path then the next genera- wanted to see. Was “” the biggest robotic road service and automaton snow- tion of robots and navigation technology will winner, or CMU’s two vehicle podium win- plows, let us generally conclude that auto- grow quickly and seamlessly into our daily ners, or any of the five vehicles that success- mated vehicles offer a good opportunity to lives. fully went the distance? Only one would relieve the pressures on our urban infrastruc- recoup a fraction of their investment immedi- ture. The immediate benefits experienced by Anthony Melihen ([email protected]) is a ately. The others will too, eventually, through the military (by placing less personnel in vul- technical communications writer at Applanix patents and established reputations as nerable positions) will quickly be introduced Corporation where he specializes in terrestrial, robotic “Centers of Excellence”. to civilian commercial applications, and the airborne, and marine applications. Louis Nastro The biggest winners, however, will ultimately infrastructure changes that soon follow may ([email protected]) is director of land products be us. The DARPA Grand Challenge may have have the biggest impact on our cities since at Applanix Corporation where he leads development been the biggest push for planners and engi- the inventions of the elevator or the electric in terrestrial imaging and navigation systems. neers to proceed with automated transporta- trolley. In all application models, many lives More information about Applanix can be found tion in ten years, perhaps spurred hardest by will be saved. at www.applanix.com. the need to comply with a congressional mandate for a third of U.S. military vehicles Data Gluttons to be unmanned by 2015. As a result, the A unique supply-side role awaits the survey- descendants of DARPA may be around soon- ing, mapping, remote sensing, and GIS er than later. communities within the future operations

Latest News? Visit www.geoinformatics.com March 2006 45