® BuscaLegis.ccj.ufsc.br

Special Edition of The Robotics Column

 

The DARPA Challenge:
What the Autonomous
Vehicles are up Against

 

 

--------------------------------------------------------------------------------
By Paul E. Grayson

--------------------------------------------------------------------------------

It Takes a Village...

Our racecar company is actually a collection of other businesses working together to make our participation in the race possible. It is only through the combined efforts of many people that our raceteam exists. I would like to thank PC AI Magazine www.pcai.com for making access to events and people possible.
Allow me to introduce myself. I am Paul F. Grayson, Team Leader of an international all volunteer group of scientist and engineers who have formed a company called American Industrial Magic, Llc to enter the Defense Advanced Research Projects Agency's (DARPA) Grand Challenge Race (see PC AI Magazine 16.5). DARPA has started calling us Team AV because we have entered three vehicles, each beginning with the letters AV for Autonomous Vehicle. They are:

AV ANDREA MORGAN - a sand rail that looks very similar to the picture of NRaD's tele-operated sand rail, (see figure 1)
AV SYDNEY BRINSTOW - a hummer shown here in a sales brochure photo (see figure 2) and
AV WENDY DARLING - a two and a half ton cargo truck shown here as an ink drawing (see figure 3).
More about our activities are available at the website http://aimagic.org, sponsored by WDWeb Company in Traverse City. At last count there are 68 members on the team, each a specialist in one or more areas of expertise required for the project. Three people joined the team today, two computer network wizards and one expert mechanic. I joined the team so I could write about the AI technology, the inner workings of the team and the interaction with DARPA - to bring you the inside story (Editor's Note - Paul Started our PC AI's Robotic Column). It also looked as if it was going to be a lot of fun.
This past year, since I joined the team, certainly has been busy. I had not tried Red Bull energy drink before I started - now most of the team runs on it. The can label indicates it is made of large amounts of vitamins, and going by the teams performance, it certainly works. Being an all-volunteer team, we all have regular day jobs and we spend what time is available on this project. The time left before the race, March 13, 2004, will be measured in long hours.

Some of the Activities

There is no typical day in a research and development project such as this. It is for instance Sunday evening and I am still finding clocks and watches to set back an hour for the time change that happened early this morning. The readout from the atomic clock at the bureau of standards that we have here changed itself like it was supposed to do. All clocks should do that. The blinking 12:00 on the VCR should know how to set itself; it is after all a radio receiver. The computers on the network checked in one by one to say that they had changed the system clock to reflect the time change.
All that was going on in the middle of a very long phone call from several states away. Fred Galliger of D F Galliger Construction Company in Mertztown, PA called in response to a letter I had written to the editor appearing in the Dec 2003 Military Vehicle Magazine asking for someone to donate an old Army cargo truck to send on a dangerous mission; a mission to demonstrate the technology the US Army desperately needs which will save soldier's lives (eliminating drivers in war zones) - one last mission for a retired Army truck - the DARPA Grand Challenge race (www.darpa.mil/ grandchallenge). Fred called to tell us he was donating the truck and wanted to hold a fundraiser for us. The fundraiser he is organizing, which includes appearances and speeches by government officials, is going to be a picnic style party in Mertztown Nov. 15, 2003 complete with a retired military vehicle convoy. It will be a big send-off |for AV WENDY DARLING as it heads off for Traverse City, MI where additional modifications will be done in preparation for the race. Exotic Cutting Tools (www.exoticut.com) has supplied the precision measuring tools and cutter bits we will be using. They are quite shy though and will not be appearing in the movie being made documenting our preparation for this race.
AV WENDY DARLING, is to be outfitted, using tools donated by Fastenal, with a state of the art guidance and control system running a special blend of software written using tools donated by NeuroDimensions (www.nd.com), The system must be modified to make the 300 mile run from Los Angeles to Las Vegas. Tripp-Lite provided the power conditioning equipment to run the computers during the trip. During the race, the vehicles will be traveling on their own, watched from a distance by Department of Defense range safety officers. The all wheel drive 6 x 6 truck, AV WENDY DARLING, must find its own way from one race route GPS waypoint to the next. The terrain it has to deal with are shown in photographs on the DARPA web site (see www.darpa.mil/grandchallenge/media_images.htm).

The DARPA Challenge

When DARPA started showing photographs of the hazards along the planned route to the 423 people in the audience at the pre-race conference Feb. 22, 2003, there was a murmur in the crowd as people whispered, "yea, we can handle that." As the slide show progressed, the crowd grew increasingly agitated. When they viewed the truck fording the stream (see figure 4) the crowd jumped out of their seats and started complaining to the judges that there was no way that driverless vehicles could ford a stream. They were saying that having to deal with water in a desert race was asking too much.
As people were throwing their hands up in the air in a gesture of defeat, I was puzzled - the truck in the picture made it through this water. When the judges pointed out the steep bank just past the water in the next slide (see figure 5), and that this bank would be a significant obstacle, the crowd did not even hear it; they were still complaining about crossing the water. I have seen jeeps ford streams with a pipe hooked on the air intake, with only it and the driver's head showing. My amphibious warfare training included caterpillar tractors with twenty feet of air intake extension on them. Personally, I would not want to ride off the ramp of a Landing Ship Tank (LST) and hang on to the controls of one of those, fifteen feet under water, but I met people who have done that. There is always plastic foam to float a vehicle across the water - one of the teams is planning to do that.
The terrain pictures show a number of other issues. It is not always possible to drive directly from one waypoint to the next - in fact, there is a cliff and if the vehicle takes the most direct route, it will probably not be able to finish the race. There is also a switch back disappearing around the hillside. The vehicle will have to determine that this is where it must go to reach the waypoint at the bottom of the cliff. This is a very tough navigation and problem solving challenge - all done without human intervention.
There is an area of tire slicing sharp rocks to drive over, and a low railroad underpass (see figure 6). The rules state that although there is no limit to the size of the vehicle, it must be able to draw in its tentacles and ooze through like a squid, if it is bigger than ten feet by nine feet.

Evolution of a Design - Entering the Realm of Darwin

When you examine the pictures carefully, the desert is mostly the consistency of a gravel road. What is between the tiny gravel pieces is another problem - talc like powder. Light beige in color, it hangs in the air when disturbed - possible vision issues. Whether dust is going to be a problem depends on when the last time it rained. At the scheduled time for the race (March), the Mojave Desert is its rainy season. Therefore, temperatures are not as hot, but flash flooding is a possibility. If slightly damp, the dust will stay on the ground – or may stick to camera lenses.
This brings us to the plans for our three vehicles (see figure 7 for a control block diagram). Using two GPS's located as far apart from each other on the vehicle as possible, yields better position and orientation information with respect to the on board digital maps (see figure 8). Figure 9, which shows the basic sensor layout, does not show the video cameras mounted on the vehicles - simple USB web cams. In the past, these have been the only sensors beyond GPS on these types of vehicles. Unfortunately, if the vehicle is not the first one, then dust may obscure the view completely. Wind direction may cause a single cars own dust cloud to render optical system - laser, video, etc - unusable.
We added two radars to allow driving in dust clouds, each scans 270 degrees giving a total of 360-degree coverage around the vehicle. Although the radar can see wire fences, fence posts, rocks, and other obstacles, it cannot see a plastic snow fence. The rules state that the course may use these plastic fences to mark areas where vehicles may not travel. Therefore, we added high frequency sonar to see the snow fence and other radar transparent objects.
Challenge guidelines indicate that GPs alone may be insufficient to have a vehicle reach the finish line. They indicate there are places or times when GPs will not be available. I suspect they are going to turn it off part way through the race just to see what the vehicles do - after all, they are the Department of Defense and can do things like that - it is called selected availability. Therefore, we need a dead reckoning system. In our case, we have elected to use a precision inertial navigation system. When I was in the submarine business, we used Ships Inertial Navigation System (SINS). The inventor tested it by putting it in a milk truck (it was rather bulky back then) and drove it all over the hills of San Francisco. After driving it for hundreds of miles, it was very precise - almost exactly what we want to do.

How Precise Does the Navigation and Guidance System Need To Be?

AV WENDY DARLING is eight feet wide and portions of the race route are only ten feet wide - leaving one foot on each side before going out of bounds. Some GPs systems boast 1 cm accuracy when stationary but moving their accuracy drops to 15 cm. That leaves us about 5 inches on each side, which is very tight. We will have to connect everything together and test this part closely. The minimum average speed is approximately 30 MPH to finish the race within the time allotted. Therefore, for every hour spent at 5 MPH another hour has to be spent at 55 MPH. For everything to work at 60 MPH over rough road / terrain, will be a significant challenge.
The race route (www.darpa.mil/ grandchallenge/route_def.htm) is approximately 1,000 GPs waypoints and the boundaries are set as an offset from the line that connects consecutive pairs of waypoints. At some place in the race the width of the corridor is only ten feet wide, leaving only one foot on each side of AV WENDY DARLING before the vehicle would be declared out of bounds and disqualified. Our initial approach was without the vision system since radar and sonar handle the obstacle-finding task. The race officials pointed out that it is important for systems to stay on the road and that radar and sonar would not show the painted road markings the vehicle needs to stay within. Since the road has no curbs and the shoulder is graded to the same height as the pavement, a vision system must see the road edge, which is only a color change - pavement vs. desert sand color.
It is necessary to avoid going out of bounds since there is only one penalty now - disqualification. A foot note in the rules says that if your vehicle starts to go out of bounds and does not respond to the range safety officer's electronic stop signal, the vehicle will be stopped by any means necessary and that some damage may be done to the vehicle. When discussing the race with people this is where I tell them about the helicopter gun ship. This would no doubt be something the news would show over and over again... close ups of Britten Banners' vehicle-wrap art filled with sponsors logos being perforated to look like a pegboard.
Back when I was in the missile testing business, the range safety officer gave us destruct packages of explosives to launch. If it went off course, the range safety officer blew it up. I suggested to the DARPA judges that they do the same thing. If they have to hit the electronic shut down button for the vehicle (not the explosive) - it is disqualified and out of the race. On second thought, a documentary company is filming us preparing for the race and if our vehicle is disqualified, it might as well go out in a spectacular orange blast - it would look good on film.
However, the judges pointed out that the land use permits require that all pieces be removed and that blowing it to bits, while being a dramatic end to 18 months work, would be very tedious to pickup. Riddling the vehicle with 50- caliber gattling gunfire would likely stop the vehicle with the fewest number of pieces to pick up afterwards. Plans are to call out the Army reserves to help guard the course and prevent vehicles from wandering away from the race.

The e-stop has three modes of operation:

Enabled - Normal operations
Shutdown - used when the vehicle is disqualified.
Normal stop - used by the judges to temporarily halt the vehicle for any number of reasons, and then allow it to proceed.

The race officials supply and control the e-stops. Easily reached manual shutdown switches are also positioned around the vehicle. The trick is to create an easy to operate switch that will not be activated by shock or collision with other vehicles while racing. Dr. Vincent Cornellier, who has a strong personal interest in Artificial Intelligence, has given the team a cash donation to spend on outfitting our vehicles with these and other necessary items.

The Race Environment

A drawing two or more hours before dawn on race day will determine vehicle starting position while the desert is still very dark and cold. There will be a meeting of team leaders two hours before dawn and the route information disk handed off to the vehicle crew chief for loading into the vehicle. As soon as the e-stop halt signal is removed, about 06:30 local time, all vehicles will start their flashing yellow light and begin to emit their 113 dB intermittent noise for five seconds before starting to move. They will keep it up for the next ten hours (minus a 15-minute wait at the reviewing stand at mile 180 of the course. Pit crews had better wear double hearing protection for this part of the event. Sonars will be filling the air with acoustic energy too. Radars, lasers, all manner of chaos will be going on, marking the start of the race.
More later!

 

--------------------------------------------------------------------------------
Paul F. Grayson is Chief Engineer at American Industrial Magic. His life long interest in robots led to his traveling around the USA to meet the people involved in robotics and his founding of the ROBOT CLUB of Traverse City, MI. www.wdweb.com/robotclub/index.asp. Mr. Grayson maintains an extensive collection of technical papers and books on the subject of robots and is frequently asked to speak on the subject. He can be reached at (231) 946-0187 or pgrayson@traverse.net.

 

Fonte:http://www.pcai.com/Paid/Issues/PCAI-Online-Issues/17.3_OL/New_Folder/DOH912/17.3_PA/PCAI-17.3-Paid-pg.46-Art5.htm