The DARPA Grand Challenges

vehicle automation in the nineteen eighties and nineties. But that work would only go as far as the sophistication of the technology of the time would allow, and funding for AI work had gotten pretty darn scarce in the eighties and nineties. So while he and his team were prepping for their ninety four demonstrations that were so impressive, there was similar work going on in the United States. From

to nine. Research teams took advantage of six hundred fifty million dollars of funding created by the US government through the Intermodal Surface Transportation Efficiency Act. That act was meant to support planning for several transportation projects, uh including the creation of new rail lines between various cities. At the autonomous research was really just one piece of this larger legislation.

And also just to let you guys know, in case you're curious, many of those projects did not receive enough funding to really accomplish their goals, so they never really self fruition. It's one of the downsides of government projects is that the budgets are too low. Nothing of real import tends to happen. People might get paid by some of that money, but the actual projects don't necessarily, you know, happen.

On the driver list car side, nine organizations began to work on technologies to make driverless cars a reality, but the deadline for meeting those goals in a suitable demonstration was and from what I can tell, not a whole lot of progress was made in that time, or at least not enough to end up with an impressive demonstration. That marked a point in history that we would say, this is really important in driver lest cars. But we're continued on the various systems that would be necessary to

make driverless cars. But you would argue that driverless car research and development as a whole, as a as a goal in of itself, was kind of shelved for a while. There were people working on various parts of the problem. Most of them were working in fields that ultimately would prove important to the development of driverless cars. But again, they were doing this for other reasons, and it wasn't necessarily as uh an in goal of creating an autonomous vehicle.

Much of the excitement and money around the concept had died down by the mid to late nineties. Some of those related research projects would end up growing out of a US backed R and D project called the Strategic Computing Initiative or s c I now that actually got started in ninety three. The goal of the initiative was incredibly ambitious. That's a very kind way of putting it.

I'd actually say it was unrealistically ambitious. It was to develop the technology to create a machine learning system capable of running quote, ten billion instructions per second to see here, speak and think like a human. The degree of integration required would rival that achieved by the human brain, the most complex instrument known to man. End quote. That was actually written in an account by Alex Rowland and Philip Shiman for the m I T Press about the s

c I project. The goal was to achieve that in just a deck aid so starting in nineteen eighty three, the plan was that by ninetee, we're gonna have a computer that can think and and experience like a human can. Wowsers. So it's pretty obvious by the late eighties that this was not gonna happen. They were not going to hit that goal, and in retrospect, we would probably call this the result of hubrists. The human brain is far more complicated and technology is far more limited than we gave

either credit for back in nineteen eighty three. Nevertheless, the Defense Advanced Research Projects Agency or DARPA and the Department of Defense, which is the department that oversees DARPA, poured about a billion dollars into funding various programs throughout the

United States in an effort to achieve this goal. And while we did not get a computer that thinks like a person out of this whole process, many influential computer scientists and research projects were able to advance our capabilities and understanding through their work which got funding from this project. So we didn't get what the project was aimed to produce,

but we did benefit from it. Some of that work would become really important for the next big DARPA initiative that I'm going to cover, and that is the Grand Challenge. The history of the Grand Challenge dates back to two thousand one. At that time, the United States Congress had its own challenge for the various branches of the U. S. Military. The US Congress said, we want you to develop the technology necessary to allow one third of all military ground

combat vehicles to be unscrewed by two thousand fifteen. In other words, to have autonomous ground combat vehicles one third of all of them. The purpose, obviously was to keep soldiers out of harm's way as much as possible. That if you could have these ground combat vehicles operate autonomously, then if one gets destroyed, that's a huge amount of money gone down the drain, but no one dies, at least no one on your side dies. Achieving this goal

would also be super difficult to do. Technology just wasn't where it would need to be by itself. The defense contractors that DARPA would work with on these sort of solutions weren't making the progress necessary in order to meet that two thousand fifteen goal, and DARPA recognized this early on.

They said, this is just not gonna happen. So in the final report for the two thousand four Grand Challenge, which was the first of the three Grand Challenges around autonomous cars, the logic that the agency laid out to justify this challenge was this, while there have been a number of significant technical breakthroughs leading to robust unmanned air vehicles that US forces used today, progress in unmanned autonomous ground vehicle technology has not occurred at a similar rate.

Vehicle operations in a ground environment are a much more difficult challenge due to terrain, man made obstacles, and weather that's just scratching the surface. They actually were pretty you know, generous in that regard, because as it turns out, there are a lot of other factors that make this a really difficult problem. And I think one other thing quick tangent that I think is important here is DARPA pointing out, yes, we came up with autonomous systems for aircraft ages ago.

We have lots of them. We have unmanned drones that we can fly, but we haven't really done that with cars. I think that's also a good reminder that technology does not all progress at the same accelerated rate. We have more is law, which has kind of conditioned us to think about our technology advancing rapidly over the course of every two years. But that doesn't apply to every technology.

It applies specifically these days to computational power, and originally only dealt with the number of transistors you could fit on a square inch of silicon wafer. So this is good to remember because I think a lot of futurists just kind of apply More's law to all technologies and just assume that everything is accelerated at that saint or everything is moving at that same accelerated rate. Back to

autonomous cars. Jose Negron, who worked at DARPA at the time of the Grand Challenges, went to the director of DARPA with an idea, and he said, there are a lot of people out there who could potentially contribute to the advancement of technology we're going to need for autonomous cars.

But we normally would work with them because they're independent innovators, They work with smaller groups, they aren't part of defense contractor companies, and so they don't often get a chance to do any sort of work for the Department of Defense. But we could take advantage of them if we are if we give them the chance to participate, we could benefit from that and they might create the breakthroughs that

we need to meet that two thousand fifteen deadline. And so, after a couple of years of funding various research projects, the then director of DARPA was a guy named Tony Teather. Anthony Tather announced a new initiative and it was for an open challenge to any teams that wanted to participate. Anyone could apply to be part of this challenge. Specifically, the challenge was to develop autonomous vehicle technology and incorporate it into a vehicle that could traverse a one forty

two mile course without human intervention. So from the start of the course to the end of it, the vehicle would have to be operating under its own abilities, its own power. The team to complete this course in the shortest amount of time would take home a million dollars. Now, initially, Anthony Tather was skeptical that this announcement was going to draw much interest. He said, we'll get a shot, but I don't think very many people are going to respond.

DARPA scheduled a kickoff event at the Peterson Automotive Museum in Los Angeles, California for anyone interested in participating. On the day of that event, Anthony Tather arrived at the venue a half hour before the scheduled time to start, and he discovered there was a line wrapping around the block of people who were interested in participating. According to the final report for the two thousand four Challenge, more than four hundred people showed up to that first conference.

The race course for this two thousand four event would end up being incredibly challenging. A guy named Salve Fish actually designed the course and he made it really tough. Vehicles would travel through the desert, including trips up and down hills and through switchbacks. Sometimes the road would narrow down to about ten feet wide. There are points where the road would drop off several feet to one side

or the other. The course crossed railroad tracks. There was the chance that the cars might encounter animals along the course. DARPA had some professional drivers actually go down this course after it had been established to kind of give their estimation of it, and they said it was fairly challenging even for a professional trained off road driver. Any autonomous car that was to complete this course was expected to do so within ten hours of starting, so there's a

ten hour time limit essentially from start to finish. So why did they make it so challenging. Well, keep in mind, the ultimate goal of this and the following challenges was to encourage the development of technology that would allow the US military to build autonomous military vehicles. And the US was and still is, heavily involved in operations in the Middle East, and so the course was partly designed to mimic the conditions that military vehicles might regularly encounter in

that part of the world. And while on a test we might say, while that's really challenging, in a real world scenario, we don't really. We can't change the parameters. It's it is what it is. So it needed to be really hard. It needed to be hard also to spur on that innovation. The course would begin in Barstow, California, and it would end in prem Nevada, passing through the Mojave Desert. All right, I'm gonna talk about how this challenge unfolded, but first let's take a quick break to

thank our sponsor. The teams had just one year to develop, build, test, and complete their technologies. So, like I said, this was an enormous challenge. Competing teams included universities, there were research facilities, There are some company teams. There were even some high school teams that applied. DARPA received one hundred six applications. Now, not everyone who attended the kickoff event went through with the step of applying, and the rules did restrict who

could participate. They said US federal government organizations could not participate in this event. Federal employees could participate, but only as a private citizen on a team. They could not

represent a federally backed team. Federal funding wasn't allowed either, and teams were prohibited from using government owned equipment, with one exception, which was that if the government were to offer equipment to all teams, that's fine, but no team could just take advantage of government owned equipment if they were the only team that had access to it. DARPA set up a website for the competition that included a

forum and which interested parties could post questions. They could share advice, They could share strategies for how they were going to solve really hard problems like obstacle detection, path finding, position location, the control software that would be needed for the vehicle itself. Teams could also help each other find sponsors, to help get money and funding to cover expenses, and it sounds to me like it was a fairly collaborative

space despite the competitive nature of the objective. The general strategy for most teams was to pair several technologies together to create an autonomous car, so they weren't necessarily developing all these tools themselves. In some cases, they were making use of existing tools, so a lot of them would use stuff like a GPS receiver. I've talked about GPS extensively not too long ago, so I'm not going to

go over that again. Also optical sensors, so essentially cameras and similar sensors, laser range finders, lidar systems, things like that. Some teams spent most of their time developing the computer programs that would make decisions based on the incoming data from these various sensors, such as whether a vehicle should accelerate to climb a steep hill or to break to avoid going off the edge of the road, or to

steer in a particular direction. One team created a system that used a voting mechanism to decide on what to do at any given point, and it would wait various factors in that decision making process so that the computer could arrive at what it thought was the best course

of action for any given set of circumstances. DARPA chose twenty five of the teams after reviewing their technical applications, and immediately upon looking at the technical applications, they were able to select nineteen teams by itself, They said, all right, these nineteen really have a solid application. For the other six.

They visited several of those teams for site visits, kind of to get an idea of how far along each project was, how realistic the chances were that the team was going to be ready by race time, and generally just get a gut feeling for who had a real chance to to do something special at this competition. The teams were invited to a qualifying event at the California Speedway in Fontana, California, and of the twenty five that were selected, twenty one showed up. Four teams did not

make it to the qualifying event. The qualifying event tested vehicles in many ways, including responsiveness, speed, safety. They went through some areas that would simulate what the cars would encounter on the actual race. Only seven teams were actually able to navig Gate the full qualifier, but DARPA selected another eight on top of the seven that made it through. They were close enough for consideration, and so the on plus teams that had applied the competition was narrowed down

to fifteen teams. For the big race itself, DARPA had some other special rules. One was that each team would not be given any information about the course until about two hours before the actual start time. Another was that a human driven chase vehicle would follow each of the autonomous vehicles, and inside that chase vehicle there would be a laptop operated by a team member that would allow

them to activate a kill switch. So if an autonomous car or motorcycle more on that in a second where to go nuts, start careening off course and driving out of control, a human in the chase vehicle could hit the kill switch and that would shut down the target vehicle before any real calamity could occur. The cars weren't meant all to leave at the same time. It wasn't like that kind of race where they all lined up

and then the flag dropped. There was a staggered departure time UH, and it was staggered by several minutes so the goal was to have the fastest time from start to finish, but not necessarily to physically race against another vehicle in real time, although there was the potential that a car could catch up with whichever vehicles were in front of it and pass it that way, that was

a possibility. DARPA told CNN that about five people were working on the race in total, some of them were volunteers, some of them were DARPEST staff members, and that the race also employed several wreckers tow trucks. Essentially, so if vehicles got stuck, a wrecker could come in and pull them out of the way, and that vehicle would obviously be uh. It would be eliminated from the race. Uh and whoever pass that finished line in the shortest amount of time would walk away with a cool one million

dollars in prize money. Now, one of the participants in this race, just as a moment of interest here was a guy named Anthony Lewandowski. And I've talked about Lewandowski in tech stuff. I think it was a year ago when I was talking about Lewandowski. His approach was unique in that he brought a self balancing motorcycle to the

event rather than a four wheeled vehicle. Lewandowski became somewhat infamous for departing Google he worked in their autonomous vehicle department and then going to work for Uber and then allegedly bringing along with him some proprietary information from Google. So that was what I was talking about the last

time I chatted about Lewandowski. The fifteen teams, in order of their start time were the Red Team from Carnegie Mellon University, PSI Autonics two from Thousand Oaks, California, Team cal Tech from California Technical University, Digital Auto Drive or Dad from Morgan Hill, California, Virginia Tech which is from Virginia, Axion Racing from Westlake Village, California, Team Cajun Bot which might have my favorite name there from Lafayette, Louisiana, Team

Endsco from Falls Church, Virginia, Team SIMAR that's see I M A R from Gainesville, Florida, and Logan, Utah. It was a joint project the Palos Verdes High School Road Warriors from Palace Very Days Estates, California, PSI A Tonics one from Thousand Oaks, so they actually left the starting line well after PSI A Tonics too uh. Team Tera Max from Oshkosh, Wisconsin, Team Tara Hawk from Guardina, California, the Golem Group from Santa Monica, California, and the Blue

Team from Berkeley, California. At six thirty in the morning on Saturday, March two thousand four, the Red Team's vehicle called Sandstorm so It's a Sandstorm from Carnegie Mellon became the first of the driver lest cars to tackle this course. Now, it would get about seven and a half miles down the road, but then it got stuck while trying to

navigate a tight switchback. It was the most successful of all the vehicles, seven and a half miles out of a forty two So out of the fifteen cars that made it through to the competition, two of them withdrew prior to the start of the race. That would be Tara Hawk and Blue Team. Four of them didn't make it out of the starting area because of various problems ranging from steering abnormalities to colliding with wall, which I

guess in its way as its own steering abnormality. Then you had Team Ends Goes vehicle which made it point two miles before it flipped over. Team Simar's car got tangled up by wire. Half a mile in the Golden Group's vehicle got stuck on an incline and couldn't provide enough throttle to overcome it. Team Terra Max's vehicle kept detecting bushes and eventually just stopped moving forward after it thought that there were bushes everywhere. I guess wouldn't budge.

Team cal Tex car went off course and through a fence and could not get back on course, and so it was disqualified. So none of the vehicles were able to complete this on two mile course, not by a long shot. No one collected the million dollars. But DARPA wasn't about to give up there. For one thing, the agency had created an incredibly ambitious challenge, a really hard one, and for another, the teams were learning from their mistakes, and so DARPA chose to announce a second challenge that

would take place in two thousand five. It would double the prize money to two million dollars and the ameters of the challenge would change slightly. Now, according to the final report for the two thousand four challenge, the agency quote believes it prudent to continue with the prize authority approach and hold a second Grand Challenge for autonomous unmanned ground vehicles in two thousand five. The prize authority approach is meeting the goals of attracting new talent with new

ideas and accelerating advancement in robotic vehicle research. Without the Grand Challenge, it is doubtful there would be much progress without substantial new investment in accelerating research on autonomous ground vehicles that could traverse difficult terrain at militarily relevant speeds end quote. So how did that turn out? Well, I'll tell you. First thing, get a drink of water, and we're gonna thanks O our sponsors here. As Anthony Tether would say later, the thing to remember was the two

thous and four Grand Challenge was something new. It had never been done before, and it was an engineering challenge open to anyone, and teams that were able to meet the selection criteria were able to participate. So while there was disappointment when no car was able to make it over the big hill that Sandstorm got stuck on, it

didn't diminish the excitement for a second Grand Challenge. Returning teams were energized by the need to outperform their previous attempts and new teams came forward inspired by the original challenge. So while it could have been the needle that would deflate the autonomous balloon could lead to another artificial intelligence winter. Like I said in the last episode, it actually got

more people excited about those possibilities. So in August two thousand four, DARPA held another participants conference for parties that were interested in either participating or sponsoring a team. Now, remember it just held the Grand Challenge in March. Now it's August and they're ready to talk about the next one.

The new challenge was scheduled for October eight, two thousand five, and the rules allowed for flexible starting time, So if the weather was bad or other conditions were such that you couldn't really start in October eight, you would start the next day or the next up to October eighth, two thousand five. If it kept missing it up to that point, the whole thing would be canceled. Now, according to the rules, they said the route would be quote

no longer than one miles end quote. I guess that was a relief since no one got past seven and a half miles in the first one, and also said the route would include quote paved roads, unpaved roads, trails, and off road desert areas. Examples of obstacles include ditches, berms, washboard, sandy ground, standing water, rocks and boulders, narrow underpasses, construction equipment, concrete safety rails, power line towers, barbed wire fences, and

cattle guards end quote. The rules also stated that DARPA could introduce obstacles onto the course. They could purposefully put some obstacles in the way, but that the route would be wide enough so that a vehicle could bypass the obstacles without going off course. So they weren't going to block a road entirely, but they would at least partially

block the road in certain locations. Now, if you want proof that the two thousand four challenge did not discourage IT participants, you just have to look at how many people and teams applied to be part of the two thousand five challenge. There were a hundred six people or teams that applied in two thousand four and two thousand five it was one hundred nine five teams, nearly twice as many that showed interest the year before. Out of those on DARPAT selected one eighteen teams for site visits.

Out of those on eighteen, they chose forty that were selected to go on to the national qualification event, and then they picked three more alternate teams that were added in August two thousand five. Now, like the two thousand four qualifying event, this one was designed to sim relates some of the conditions for the final race, with the goal of reducing the number of competitors to the top

twenty teams. All but one of the teams passed the initial technical inspections to make certain the vehicles met all the safety and performance parameters, and according to a news release from October two thousand five, twenty two robotic vehicles were able to get through the obstacle course designed to simulate the final race. DARPA actually would select twenty three

teams to compete on the day of the race. They included returning teams like Cajun Butt, cal Tech, and Team Dad, and there were new teams to like Team Cornell, Desert Buck Eyes, the Gray Team, Insight Racing, and Stanford Racing Team. The final course for the two thousand five Challenge was one thirty one point two miles, which is about two hundred eleven kilometers. Out of the twenty three competing teams, five teams finished the full course. Only four of them

did so within the ten hour time limit. The Terra Max, which was fifth place, would finish in twelve hours fifty one minutes, so it took a little too long to make that ten hour time limit, but it's still finished. All the other vehicles experienced either mechanical or software failures somewhere along the route and we're unable to complete the full course. The winning team of the two thousand five challenge was the Stanford Racing Team and their vehicle named Stanley.

The average speed for Stanley was nineteen point one miles per hour, which is about thirty one kilometers per hour. Stanley was built on top of a Volkswagen tou Egg, which could be largely operated by an onboard computer by linking it to the vehicle's electrical system. However, they did still have electro mechanical parts to operate the steering and the gearshift, so it wasn't all done purely electronically yet.

Stanley had five lie Dar units that could gather data for the vehicle's computer to build out a three dimensional map of its surroundings and pair that with GPS information, and there was some other positional equipment aboard Stanley as well to help supplement the information gathered by the GPS receiver. The computer system operated on Lenox, so Lenox fans out there, that was the computer system that was running on Stanley.

According to Stanford, the code for the vehicle's behavior consisted of a hundred thousand lines, and it's decision making process was guided by machine learning. The team completed the course in six hours fifty four minutes, and that meant that they came in about eleven minutes ahead of second place. Second place went to Red Team's Sandstorm vehicle. Again, that's Carnegie Mellon University, and they were the one who once who got the furthest the year before. But Red Team

had also fielded another card. They didn't just put in one, they put in two, and the one was called Highlander with a one in place of the eye. I guess there can be only one, but anyway, they put forward Highlander as well. That one came in third place, so Carnegie Melon took second and third place. It was nine minutes slower than Sandstorm, so it was about twenty minutes behind the first place winner. Now interesting side note about Highlander.

Initially it was making great time, like really really good time. It was on track to win the whole darn thing. But about two hours after starting the race, the vehicle's engine began to sputter a bit, and it struggled anytime it hit a steep climb, and it was never able to get up to full speed even on a flat or or declined surface. So the Carnegie Melon team was really curious. They were wondering what the heck was going wrong, and they tried to figure out what happened at the

end of the race. They weren't really able to nail it down at that point. They checked the fuel. The fuel was fine. It wasn't contaminated or old or anything like that. The oil was fine. The engine appeared to work just fine from a cold start, and it wouldn't be more. It would take more than a decade before they figured out what had gone wrong, and it was

just by luck that they figured it out. Now, this is going to include a spoiler, but it's a spoiler for something that happened in two thousand seven, So I don't think it's really that big a deal. In Carnegie Mellon University was celebrating the tenth anniversary of the team's DARPA Urban Challenge win from two thousand seven, and I'll be talking about the DARPA Urban Challenge in our next episode well. As part of this celebration, they brought out

Sandstorm and Highlander from storage. The two vehicles had been stored away, but they still existed while they were doing it.

Spencer Spiker, who was part of the team working on this, was running some diagnostics on the Highlander engine it was running at the time, and as it was going through this process, he was leaning against the vehicle and the engine began to sputter, and when he moved away, the engine started to pick up again, and he realized that there was a little box that if there was pressure put on that little box, it started to cause the

engine to die. That little box turned out to be a filter that sat between the car's engine control module and the fuel injectors, and he found out that if there were any pressure put on that filter, it was causing the engine to lose power. And leading up to the Grand Challenge, there was an incident that may have

caused this problem. The Highlander had been going through a uh kind of a routine training exercise, and as it was doing so, it got too far over to one edge of a sloped path and it slid off that

sloped path and it actually flipped over. Now, apparently that accident, which appeared to be minor, it didn't seem like it had caused that much damage, had actually bent this little box, this filter in such a way that because of its contact with the engine, was causing failure under the right conditions, such as when pressure was put on it or when

it would expand from engine heat. And so the Highlander, after it would start to heat up, would start to lose performance, and it lost time during the Grand Challenge, about forty minutes worth of time according to most estimates, which would have made all the difference. It would have come in first place by twenty minutes if that had not happened. So it may have won if it hadn't

been for that accident that had happened earlier. Making things even more juicy is that the leader of the Standard team that did win was Sebastian Throne, who had previously served as a member of the faculty at Carnegie Mellon University.

He was In fact, a colleague of the head of CMUs team read Whittaker, So the Red team's Red Whittaker, the Carnegie Mellon University team leader used to work with the guy who led the Stanford team, So there was some rivalry going on there, although from what I've read, it sounds like it was all very good natured rivalry. It wasn't like it was super bitter or anything. Now, the nice thing is no one seems particularly mad that this happened. It's the sort of thing that can happen

with mechanical systems in general. And ultimately, Carnegie Mellon University would do very well in the next Grand Challenge. But that's a story we're going to cover in our next episode. Darpest statement was pretty darn positive. Quote. The Grand Challenge stimulated the creation of a new community of innovators, inventors, mechanics, computer scientists, engineers, and students who typically have not been

involved in defense related activities. The camaraderie and competitiveness that have been the hallmark of the Grand Challenge since its inception demonstrates that America's heritage of ingenuity and resourcefulness is strong end quote. Now, in our next episode, we're gonna look at the Urban Grand Challenge of two thousand seven and how these competitions lead into the autonomous car environment

that we are in today. And we'll also talk about some of the most difficult challenges we face, both technical and ethical. And we'll also talk about the pros and cons of driver list cars in our upcoming episodes, so I hope you look forward to those. It's been a lot of fun kind of going back through the history and seeing what has led up to what we see today.

It's really fascinating to me to see the the transition away from a world where all of the autonomous nature of cars is built into the infrastructure surrounding cars to one where it's largely built into the vehicles themselves, and mostly because the Department of Defense needed innovation in that space in order to meet a deadline to have one third of all ground military combat vehicles automated by two fift we'll talk more about that as well in an

upcoming episode. For the time being, if you guys have any suggestions for future episodes, or you've got any suggestions for people I should have on the show, or maybe there's just a topic that you've always wanted to know about Send me a message. The email address for the show is tech Stuff at how stuff works dot com. Or pop on over to our website that's tech Stuff podcast dot com. You'll find other ways to contact me there.

You can reach out on social I'd love to hear from you, and don't forget to go to our store. It's at t public dot com slash tech stuff. Every purchase you make goes to help the show. We've got lots of cool designs on there. I'd love you to check those out. I think I'm really proud of them. I'm really happy with the way they've turned out, and check them out if you If you like something and you buy it, like I said, some of that money goes to help our show and we greatly appreciate it.

And I'll talk to you again really soon for more on this and thousands of other topics because it how stuff works dot com.