DARPA (Defense Advance Research Projects Agency) is a part of the U.S. government tasked with bringing new and revolutionary weapons to the war-fighter. One of its goals is to develop a vehicle that is capable of traversing areas with existing unpaved roads and trails without requiring the guidance or intervention of a human driver or controller. Such so-called “autonomous” vehicles would have to be capable crossing areas of terrain similar to the U.S. and Iraqi deserts following only tracks of GPS-based coordinates to get from their starting point to the finish. These vehicles need to incorporate enough sensors and “smarts” to be able to “read” the terrain in between these points and avoid natural obstacles that would disable or destroy the vehicles. The eventual goal is to robotically re-supply troops in the field without exposing drivers to hostile enemy action. As a matter of fact, the Department of Defense (DoD) has since established a goal of having at least one-third of the vehicles in the military autonomously guided by the year 2015.

DARPA had first approached a number of aerospace contractors with this task, and met with little success. Eventually, they realized that they would have to try a new and different approach if they were going to be able to realize the goal set for them by the DoD. They decided to capitalize on the American entrepreneurial spirit and challenge smaller, less-encumbered groups to develop a vehicle that could accomplish this task.

While the capability to read and react to changing terrain comes naturally to a human rider or driver (probably because he is a product of generations of evolution and experience where his very survival depended on accomplishing it), the entrants in the 2004 DARPA Grand Challenge found this to be a very daunting challenge. As you may recall, DARPA offered a one million dollar cash prize to any of 106 teams entered that autonomously traversed some 150 miles of Mojave Desert (from the Slash-X near Barstow, California to State Line/Primm, Nevada) in the shortest period of time (but no longer than 10 hours). It turned out that just getting out of the Barstow area proved to be such a challenge for the 16 that eventually competed, that the most successful team managed to make it just seven and a half miles.

Terra Hawk staged for the first run	Terra Hawk takes off on first of three runs

Not to be deterred, DARPA doubled the prize to two million dollars, and promised to stage the event again the next year. Teams interested in competing had to enter by February 1, 2005, and then be subjected to a number of qualifying tests and exams before the final race on October 1, 2005. This year, the number of teams competing ballooned to 195, with entries from 37 states and 3 foreign countries, including 35 university-based teams and 3 high school entrants.

The first qualifying event fell on March 11 when teams had to provide DARPA with detailed information about their entries, including a vehicle specification sheet and a video demonstration of their vehicle in action. In the case of the Terra Hawk, a Northrop Grumman-sponsored entry which competed in 2004, DARPA opted to send an inspection team to examine their facility, and personally view the vehicle traversing a closed course complete with obstacles placed by the reviewers. (While Northrop is supplying all of the materials and purchasing computers, sensors, etc., they are not paying the employees who must volunteer their time. In return for providing the front money to Terra Engineering, Northrop will own the rights to all of the intellectual property that is the result of this effort.)

Two PhD’s from DARPA, Dr. Gary Graham and Dr. Leo Christodoulou, showed up at the test area used to demonstrate and develop Terra Hawk (a vacant field in an industrial complex in Dominguez Hills). After measuring the course and verifying that the width and other parameters met DARPA requirements, they placed two 32-gallon trash cans on the course to simulate obstacles that Terra Hawk would have to deal with in the actual race.

Even though it was a Thursday afternoon, a dozen or so Northrop employees were present, including most of the members of the Terra Engineering team. The suspense and tension were palpable as everyone realized that a poor showing here could eliminate any chance of competing due to the crowded field of entrants. A hardware failure similar to the one that kept Terra Hawk from even starting the last Grand Challenge would kill the team’s chances for a return to the Slash-X. Conversation ceased as we watched Terra Hawk start the course, successfully avoiding the first obstacle and staying within the course limits. It seemed to steer directly down the center of the course, except where it had to turn to avoid the obstacles.

Grand Challenge Demonstration loop (Each dot is a GPs coordinate or way-point)	Well wishers and workers look on at trial

Below is a film clip that is representative of the runs that were made. You can see that Terra Hawk perceives the obstacles and avoids them. The Kubota cart that is following the vehicle has a team member with his finger on the kill button in case something runs amok. Fortunately, it wasn’t necessary on the runs than counted.

Terra Hawk 11 May 2005 obstacle avoidance run.wmv
Note: Very Large - 15.9MB

This same successful performance was demonstrated for each of the required three runs, with the vehicle avoiding all obstacles, and staying within the course boundaries. The team was obviously excited and relieved that their creation appeared to be working so well. Todd Mendenhall, Terra Engineering lead, agreed to a fourth run at double the speed of the first three runs. Unfortunately, the vehicle trampled the trash-can obstacle, leaving everyone scratching their heads as to what could cause the sudden problem. Fortunately this run was optional and would not disqualify Terra Hawk from participating in the next phase of the race.

Subsequent investigation of the extensive data logs automatically taken on each trial run suggested that the navigation filter crashed at just the wrong time, with the most likely cause being Electromagnetic Interference (EMI) from the high current motor-generator and Pulse Width Modulated (PWM) motor drivers disrupting the data flowing from the sensors. One of the tasks that will face Terra Engineering is the re-routing of all power carrying lines as far away from signal lines as possible. If this fails, then more sophisticated isolation techniques such as installing chokes and shielding of power cables will be tried.

Terra Hawk plays crunch the can on the fourth & optional run.	Todd surveys the damage

After the testing was over, Terra Engineering posed for a group picture in front of their creation, and then the DARPA representatives interviewed the team, posing questions from an extensive list of scripted queries. I felt that Todd handled the questions well, effectively quelling some of the concerns expressed by the DARPA questions. One of the DARPA team asked what sort of speed Terra Hawk had used to maneuver the course, and then noted that this speed would not allow the vehicle to finish the course in the required 10 hours time. Todd noted that Terra was not yet in full racing form, and that the current gearing is way to low. It allows testing to be done at a safe speed, but would indeed leave it out of the running. Currently being developed is a “crash-box,” unsynchronized, 2-speed transmission effectively giving Terra an “overdrive” gear, and resulting is a significantly higher top speed. He also noted that the vehicle was not carrying the full complement of sensors, and felt that the planned addition of stereo-optical cameras would increase the speed and resolution of the data flow, and result in greater ground speeds being possible.

DARPA officials interview team members at test site	Back at Terra engineering shop

Development team poses in front of Terra Hawk

Autonomous operation Deployable sensor head Precision GPS waypoint following Obstacle avoidance 3-section articulated steering Active suspension Hybrid diesel-electric power	Physical Characteristics 80” wide, 146” long, 57” to 75” tall Weight: 2400 lbs 6-wheel independent electric traction
LADAR, optical, GPS antenna & computers Steering, suspension, & drive train elements

Shown above is a summary specification sheet for Terra Hawk, Along with close-up views of some of the features of this amazing vehicle. On the left is the sensor-signal processor “pod,” which will eventually contain a pair of stereo-coverage optical sensors, an Inertial Measurement Unit (IMU) (combination attitude and navigation sensor made by a sister-division of Northrop), a LADAR (Laser Radar System), a GPS antenna and associated receiver and three signal-processing computers that are the brains of operation. The picture on the right shows the sophisticated suspension and drive system of Terra, including (from left to right) one of the pneumatic actuators and spring-damper assemblies that steer the vehicle by articulating the three sections, a similar but larger pneumatic cylinder and spring-damper that allow Terra to raise and lower its running height by 18 inches (and absorb and isolate the chassis from the pounding that off-road terrain dishes out), and finally the drive belt (from a Harley Davidson) that couples the electric motor to each wheel.

At one point, Todd noted that the pneumatic system allowed for very rapid response, but also consumes an inordinate amount of power. Since air compressor failures were the root cause of the problem preventing a run in 2004, the team plans to switch from pneumatics to hydraulics which will reduce even further the amount of power consumed and improve vehicle reliability.

While Terra Engineering seems to be getting a handle on the vehicle’s mechanical design, the real challenge will be pulling together the data handling, sensor output fusion, and processing algorithms that will allow Terra to see and avoid obstacles and to choose the optimum path to connect the GPS waypoints, all while traversing the course at maximum speed. One of reasons for all of the team’s poor showings at the 2004 event was lack of time to develop and perfect the algorithms and code to accomplish this. It is a thinly protected secret that many of the top teams had effectively switched off all of the sensors except for the GPS receivers, and were hoping that the waypoints were close enough that they could get by with their off-road capabilities. Clearly, the winning team will need to effectively blend inputs from a multitude of sensors and make rapid decisions about obstacles that need to be avoided. Some of issues deal with how to process all of the data coming from the sensors, and coming up with ways of prioritizing data that represents a threat, and discarding the rest.

The two blue lines are the result of the processing algorithm and show the road in foreground and background. Presumably, Terra Hawk would steer to stay between the two blue lines and hence remain on the road, when there is one.

Road Following uses texture statistics to identify smooth terrain (indicated in blue) Reduces road detection to a manageable 20 bits sent to mapping for path cost fusion

While staying on the road is important, perceiving a hazard and avoiding it is also critical if Terra Hawk is to make the 150 miles necessary to win the race.

Video stream is processed using optical flow in selected regions
Side-looking flow-depth and forward-looking stereo
Depth differences indicate presence of an obstacle
Project the (r, θ) detection to 2D map

So while the mechanical design of the vehicles is important, the electronics sophistication which is required to replace the man in the loop is critical to winning this off-road event. It may be that the prize goes unclaimed again this year (after all, the barbed-wire fences that played such havoc with many of the entrants are still out there, and are still difficult to detect and avoid), but DARPA will be the big winner. After all, they are getting the benefit of cutting edge technical research and development, available at a fraction of the cost than if it came through the normal channels. I would guess that many of the teams themselves will spend more money than the prize they will take home, even if they get their vehicles across the finish line.

In any case, good luck gentlemen, and start your computers!

A late breaking note: Terra Hawk was selected byDARPA to be one of the 40 teams that will advance to the next phase of the testing for the Grand Challenge Competition that will occur from September 27 to October 5, 2005 at the Fontana Raceway.