Robotic road racers

Sept. 15, 2005
A fleet of unmanned robotic vehicles will soon be racing across southern California in pursuit of bragging rights and a $2 million prize.

Senior Editor

The Defense Advanced Research projects Agency, or Darpa, a think tank for DoD, is sponsoring the second annual Grand Challenge next month. It is a race that pits unmanned vehicles against each other and the terrain. And each vehicle is truly on its own. Team members cannot help their vehicle in anyway once it is staged in the starting chute. The vehicle that stays within the rules and completes the course over about 175 miles of southern California with the shortest official time under 10 hr wins its team $2 million.

Darpa hopes the Grand Challenge will spur research and development in autonomous, selfnavigating vehicles. They want to leverage this R&D to meet the Congressional goal of having a third of all military vehicles being unmanned by 2015.

To keep things honest, no race team will know the course until 2 hr prior to the race. At that time, all teams get a CD containing the 1,000 waypoints, roadway widths, and speed limits that define the course. Darpa made sure a commercial 4x4 pickup truck could be driven the length of the course. The agency will also police the course, making sure no vehicle interferes with another. Unlike last year, in which no vehicle made it even half way to the finish line, Darpa is adding static obstacles that entrants will have to detect and avoid.

Darpa has already whittled the field down from 118 to 40, and plans to narrow it down to 20 at the National Qualification Event at the California Speedway in Fontana on Sept. 27 to Oct 5. Here's a look at some of the racing robots.

The Roadrunner, a project of Stanford University, is based on a diesel-powered Volkswagen Touareg SUV modified with skid plates and a beefed-up front bumper. A drive-by-wire system was also developed by VW at its U.S. Electronic Research Lab.

Processing for navigation and control is handled by seven Pentium M computers shock-mounted in the streetlegal SUV's trunk. If one or more of the computers crashes during the race, the team is confident the others will be able to take up the slack. Sensors include a GPS, an inertial measurement unit with six degrees of freedom, wheel-speed detectors, four ladars and radar, and a stereo camera pair and a monocular vision system. They acquire data at between 10 and 100 Hz and feed map and position data to the controller at 10 Hz so the vehicle can avoid collisions while advancing along the route. A probabilistic algorithm developed by a Stanford researcher interprets the incoming flood of data. Outgoing control signals travel a thick spinal chord of wires to the electronic brake, throttle, and chain-driven steering column.

DAD (Digital Auto Drive), like most other competitors, can be autonomously driven or accommodate a human driver. This let the team troubleshoot the modified Toyota Tundra pickup while riding inside. But DAD is also street legal, so the team can drive the truck to the competition.

At the heart of the four-wheeled robot are a realtime stereo 3D-imaging system and a vehicle servocontroller. The imaging system relies on two DSP 6416 chips from sponsor Texas Instruments performing 36 billion pixel operations/sec. They are overclocked to 1.1 GHz, the fastest speed anyone has ever coaxed out of a 64xx chip. To jump-start speed even more, programming is done in Texas Instruments' Code Composer in assembler language. C and other high-level languages slow down processing.

The system can perceive, analyze, and react to 3D images over 60 times/sec, an order of magnitude more frequent than conventional vision systems. On the truck, this could translate into being able to navigate at 100 mph. The imaging system relies on two digital cameras separated by about a foot for stereovision and good depth resolution. But the exact distance between camera lenses is critical. The cameras must be aligned to within a pixel.

To navigate, images from the cameras are rendered into a 3D map using a multiple variable-width windowing technique. The system identifies objects between 15 and 875 ft in front of the truck by size and distance. Objects are presumed solid and to be avoided, except for the road.

The servosystem takes most control inputs from the vision system, but also uses D GPS, another TI DSP chip, servomotors, and an internal measurement unit (IMU). The IMU contains gyros and accelerometers to correct for gaps in GPS signals. The controller communicates with the servos via CAN bus. The servos are three rare-earth-magnet brushless motors, each rated at 20 hp, which will handle the throttle, steering, and braking. Although the vehicle relies mainly on GPS for navigation, it can also do dead reckoning if the GPS fails. So it also carries a pair of Honeywell compasses.

Looking like a dune buggy, CyberRider rolls across the landscape powered by a 250-hp propane-burning engine geared for a top speed of 55 mph, the maximum processing speed its navigation system will handle. Final gear reduction is handled by a chain drive and sprockets that let the team use large tires without overstressing the transmission and driveline. And the large tires contribute to CyberRider's ground clearance, which is a lot greater than you get with a standard SUV.

For navigation, the team originally planned to make the vehicle capable of handling almost anything — exploring the unknown, as they call it. So they were originally going to use several cameras and a stack of fast, parallel processors. Lack of funding and Darpa changing the rules to make this year's Challenge more of an exercise in waypoint following rather than a navigation and route-determination test put this approach on the back burner.

Instead, the team will use a multibeam ladar scanner (laser detection and ranging) from Sick to profile the terrain and follow the edge of any road, and a longer-distance Eaton radar to detect obstacles. The two ladars have 100° fields of view, 0.25° resolutions, and are mounted 6 and 3° down, respectively, atop a 7-ft-tall pole. The radar, with only a 12° field of view and 1° of resolution, sits about 4 ft high. It detects moving objects up to 250 ft away. The vehicle also carries an array of cameras with edge-detection software.

Tommy, named for Team Jefferson's namesake, is a rather futuristic-looking vehicle, thanks to its aluminum skin and egg shape. Underneath the shell, a gas-fueled engine from a Subaru Legacy provides the horsepower. The engine sits low in the frame to give Tommy a low center of gravity and more ontheroad stability. A larger radiator fan and upgraded cooling should minimize the risk of overheating in the 12-ft-long, 6-ft-wide, 6-ft-tall vehicle.

For navigation, Tommy relies on off-the-shelf GPS with submeter accuracy, ladar and radar, an inertial navigation system, and a few other means of checking out what's out front. Unlike most other vehicles, the software is all based on Java and runs on a Linux operating system. The major piece of software is Perrone Robotics Inc.'s X-bot (MAX), which was developed to control mobile, autonomous robots.

"We've used Max to quickly and economically build cockroach, rat, and cat-sized robots," says team leader Paul Perrone. "So we stepped up to the Grand Challenge to prove our affordable and extensible technology could be used in elephant-sized robots as well. And although Tommy isn't quite the size of an elephant, the challenge is."

If the Challenge wasn't daunting enough, the Blue Team, made up of engineers and innovators from U.C. Berkeley and Texas A&M, decided to base their robot on a motorcycle. So not only must it keep track of where it is, where it's going, and how it's going to get there, it has to stay balanced on its two wheels while it does so. Besides being more mobile than other entrants, the team hopes their efforts will pay off in a robot that can test motorcycle components in extreme conditions such as crashes.

The entire bike uses 32 separate electronic components, and only four are considered input sensors: a D GPS, a pair of high-speed 10-bit grayscale CMOS cameras for real-time obstacle detection, and a 3-CCD camera for road detection. The team rewired the entire bike to get rid of some "strange connections" and to fully understand the system so they could troubleshoot it in the field. They also repackaged the starter and power relays, inverter and voltage regulator, and engine timing and carburetor thermocouples in IP67 casings.

The biggest vehicle on the block will be TerraMax, a 425-hp military cargo vehicle built by team sponsor Oshkosh. The Medium Tactical Vehicle replacement (MTVR) is a 6x6 already used by the Marines and Navy, and is the type of vehicle Darpa envisions using in the future to make up unmanned convoys.

For the Challenge, the large truck gives the Oshkosh team a very capable platform on which to build. For example, right off the assembly line, this truck can handle 60% grades, 30% side slopes, and cross water 5-ft deep. And weather shouldn't be a factor since it can operate in everything from 50 to 125°F. It's also equipped with a TAK-4 independent suspension system that gives it 16 in. of wheel travel, enough to rumble over most logs and boulders. The suspension also reduces wear and improves braking. To make the 15-ton truck more nimble, team engineers added steering on the rear two axles.

The size of the vehicle makes it simple to add the equipment that makes it autonomous. For example, it carries two cameras for a stereovision system designed at the University of Parma, Italy. It also carries sonar, GPS, and an inertial navigation system to ensure the robotic monster truck knows where it is.

Sponsored Recommendations

From concept to consumption: Optimizing success in food and beverage

April 9, 2024
Identifying opportunities and solutions for plant floor optimization has never been easier. Download our visual guide to quickly and efficiently pinpoint areas for operational...

A closer look at modern design considerations for food and beverage

April 9, 2024
With new and changing safety and hygiene regulations at top of mind, its easy to understand how other crucial aspects of machine design can get pushed aside. Our whitepaper explores...

Cybersecurity and the Medical Manufacturing Industry

April 9, 2024
Learn about medical manufacturing cybersecurity risks, costs, and threats as well as effective cybersecurity strategies and essential solutions.

Condition Monitoring for Energy and Utilities Assets

April 9, 2024
Condition monitoring is an essential element of asset management in the energy and utilities industry. The American oil and gas, water and wastewater, and electrical grid sectors...

Voice your opinion!

To join the conversation, and become an exclusive member of Machine Design, create an account today!