Slashdot Mirror
Stanley and the Conquest of the DARPA Challenge
geekboy_x writes "Wired has a great in-depth piece on the Stanford team that won the $2 million DARPA prize. If you remember last year's disaster - with most vehicles falling off the road in the first kilometer or so - this victory becomes all the more amazing. The fact that the Stanford team used a 'tailgating' strategy is the best surprise in the article."
Also interesting to note is the fact that the major leaders of the Stanford team came from the Carnegie Mellon AI department 2-3 years ago.
http://www.asti-usa.com
There's no magazine called "WIERD", is there?
Is is practical? If the vehicle is going to travel more slowly than what passes for normal on most freeways, how it's going to avoid road rage incidents?
A Series Of Tubes - The Remix
FTA: "He liked to point out that planes had been flying themselves since the 1970s. The public was clearly willing to accept being flown by autopilot, but nobody had tried the same on the ground."
Just give us our flying cars then already, damnit!
From TFA on 7 ways cars are already robots:
"4. Lane-Departure Prevention
Nissan has a prototype that uses cameras and software to detect white lines and reflective markers. If the system determines the vehicle is drifting, it will steer the car back into the proper lane."
I've driven enough roads under construction that I would be seriously afraid that my car would steer me into oncoming traffic because road workers haven't bothered to paint over lines that were previously there.
Personally, I'd be interested in how these vehicles do:
1. On regular highways.
2. At speeds other than the 5 to 25 MPH tested.
I realize they're not built for that. I would just like to see how they do applying what they "learned" in the desert to real traffic situations.
But does it run linux?
Seriously though, they don't seem to go into much detail about the programming aspects of the robot. Of course they give some small details on what it ends up doing, but nothing about what language they used, etc., i.e. the interesting part.
The teams did well this year, but what disappoints me is that this year, many of the teams had relied entirely on laser range finders and GPS to navigate the course.
There was one entry, a motorcycle, which still ran completely on a vision system (cameras instead of sensors). Unfortunately, it did not do too well.
While the military can still use technology developed by the teams that completed the DARPA Grand Challenge, I think they could benefit even more from a vision system capable of doing the same thing. What use is a robot that can navigate a desert if it can't actually see anything?
... is that the CMU team relied heavily on extensive pre-analysis of the environment, and failed (at least in the sense that it didn't come in first). Stanford instead relied on a probability analysis of the incoming data, along with multiple technologies for different goals (lasers for short range data, video for long range data).
It seems that the DARPA grand challenge not only showed off the first realistically autonomous vehicles, but also laid to rest the idea that expert systems were the way forward. The way forward instead is self-teaching computers. Hooray for self-teaching AI overlords!
Those who can, do. Those who can't, sue.
I still think it will be a long time before we trust a computer to drive us around. Intersting that it used a 'tailgating' strategy...what happens if all the cars around it are also doing the same!
I was never that impressed with the CMU approach. All that manual preplanning was an obvious dead end. And the giant mechanically stablized gimbal was just too clunky. It didn't help them in 2004, when they hit an obstacle placed by DARPA, and it didn't help them in 2005, when DARPA moved the racecourse from California to Nevada to prevent preplanning. The Air Force colonel in charge for 2005 said preplanning wouldn't work, and he meant it.
Computer vision of the natural world is finally about to take off, after three decades of frustration. It's probably possible to do much of the early vision processing in a current-generation GPU, which may make it affordable. Look for new apps that connect to cameras and pick out items of interest. Read that paper linked above.
From Wired: The resulting liability issues are a major hurdle. If a robotically driven car gets in an accident, who is to blame? If a software bug causes a car to swerve off the road, should the programmer be sued, or the manufacturer? Or is the accident victim at fault for accepting the driving decisions of the onboard computer? Would Ford or GM be to blame for selling a "faulty" product, even if, in the larger view, that product reduced traffic deaths by tens of thousands?
It figures. A technological advance that would cut the number of traffic deaths by about 95% by taking drunks and maniacs out from behind the wheel, and preventing 93 year-old men with dementia from killing people, will be bogged down by liability issues should the robot kill someone. C'mon people! Even the best system will not prevent a fluke accident or yes, even a bit of bad code, from killing someone, but weight that against the number of road-rage infested idiots on the road now, driving at 100+ mph, swerving in and out of traffic, and I think libility needs to be the furthest thing from anyone's mind.
Just don't let Microsoft write the software.
GetOuttaMySpace - The Anti-Social Network
was when Thun explained how the vehicle was taught to drive by following a human driver and adapting its algorithms according to his behavior, gaining much better results than "force feeding" massive amounts of data artificially.
This has immediate implications not only for robotic cars - what if we took a human and strapped some positional sensors, voice recording, etc. and made a humanoid robot follow him throughout the day?
I mean how varied are our lives after all? Given the right processing power and sensors, the results could be interesting...
Again, a great achievement for a 'bottom up' approach to artificial intelligence
Google: A Patriot's Letter
The fact that the Stanford team used a 'tailgating' strategy is the best surprise in the article.
Not anymore.
I beleive this complaint is a little early. Based on the early successes shown in the desert, without people stepping off curbs in front of cars, and other urban hazards, I believe it is premature to say robot drivers will reduce automobile deaths by 95%. That prediction maybe true someday, but we're not going to see that next year. Or even in the next decade.
basically due to whatever circumstances (width of the road, start order etc) someone has to be in front and someone has to be behind - the fact that the Stanford vehicle was following another entry had nothing to do with how it was successful, in fact one could argue it put the vehicle in some danger if the lead vehicle messed up, rolled, crashed etc. It later passed the said vehicle to go on to the win - The article makes no mention of a "Tailgating Strategy" it does say that it was tailgating another vehicle for a bit before it passed it - not sure how this is any more strategic then when I drive to work in the morning - how about this winning strategy "Don't hit the car in front of you". Don't know why this bugged me so much, its actually a good read, I just don't know why this non-existent "Fact" was so prominent in the lead in. Sorry.. not enough coffee today....
Now all we need is a superstrong protective layer, a pursuit mode, and cool red lights on the front!
That's actually not true. There was no "tailgating". During the Grand Challenge, no vehicle was allowed to approach another while both vehicles were active. DARPA had the ability to remotely pause any vehicle. When vehicles got anywhere near each other, the trailing vehicle was paused to maintain separation. If the trailing vehicle was clearly faster, a pass was scheduled. All passing took place with one vehicle stationary and at a wide place in the road. Wired has this wrong.
YOU drive car!
As a participant of another DARPA team (Cornell -- our site is down), I am skeptical as to whether the winners of the challenge would be able to drive in a real world environment. In many ways the Grand Challenge was a toy problem, but this is not usually emphasized because they want to make it seem more dramatic.
First of all, no other moving objects on the course. When a vehicle was about to pass another, the one in front was paused so that the passing vehicle could overtake it. At no time did the vehicles have to deal with changing conditions.
Secondly, to my knowledge, there were no obstacles (which were promised) on the course. If someone knows differently, I'd like to hear about it. So we don't know to what extent obstacle avoidance is effective on those vehicles.
Thirdly, daylight and clear weather is one thing, but nighttime, rain, snow, etc. would significantly degrade the data.
Essentially the problem that the current vehicles solved was this:
Given a set of waypoints and a "corridor" outside which you will never have to go (so far the problem can be solved only by 10cm-accuracy DGPS), use your other sensors to avoid obstacles by moving left or right within the corridor.
Not very much like real world driving at all. And I'm not saying Stanford, CMU and the others didn't accomplish something big -- I'm just saying it's not what the Wired piece makes it out to be.
... and they really did an amazing job, however this is sponsored by the military.
So what is it going to be used for? Suicide bomber cars?
I wish more competitions (like F1 racing for ex.) were government sponsored but for discovering certain new advantages that are directly appliable in the public sector.
Sort of like community service, offering prizes to those who prove their technology and donate it as "public patent" for everyone to use.
Exactly. If this is truly a military project then the vehicles must be able to blind-reckon once the satellite information is blocked. What's more annoying is that the tailgating idea for Stanley means that someone else is truly paving the way. While a good strategy for the race, it's does not help an "auto"-nomous vehicle, for Pete's sake.
I think they should make the requirements harder next time and give more money.
Isn't Skynet finished yet? :)
One result from the second Grand Challenge that lots of people harped about was the fact that Stanley, the winning (and hence, fastest) entry, completed the course with an average speed of only about 19 mph. "19 mph?" quoted some of the the nay-sayers, "we're supposed to get excited about that?"
One thing that TFA points out, which wasn't mentioned many other places, was that the course rules stated a maximum vehicle velocity of 25 mph. Ideally, then, the fastest possible average speed for any entrant would likewise have been 25 mph. Stanley, at times, wanted and could have gone faster than that, and held back due to the rule-imposed speed limit. In that context, 19 mph is actually quite good, considering the terrain would have forced it to slow down over bumps and turns.
This idea of robots/machines emulating human behavior was discussed in Kurt Vonnegut's novel, Player Piano.
At one point in the novel, a skilled wood (or was it metal) worker was watched by a machine as he used a lathe. After recording his actions, the machine was able to exactly replicate his work, putting the human out of his job.
Most of the skilled workers in the novel were displaced by machines that were able to accurately emulate their work. As I recall, most of the people still employed with private industry jobs had to have advanced degrees. Even secretaries were required to have doctorates in order to maintain employability, even then they were threatened to be replaced by recent innovations in robotics.
Many of the unemployed people were then given state jobs, like road repair, so that they could earn some money and keep busy. That was until the rebellion started...
One could be sent to collect a downed pilot.
"This has immediate implications not only for robotic cars - what if we took a human and strapped some positional sensors, voice recording, etc. and made a humanoid robot follow him throughout the day?"
How about creating better GUI's?
BTW The "grand challenge" still didn't have widely varying weather conditions.
The article presents the history of Stanley is presented as a series of intellectual breakthroughs that I can understand, just like a lot of the pop science I read when I was a kid (and continue to read). I'm pretty sure each of these breakthroughs (such as learning from humans and assessing sensor data critically) are ideas that have been around in AI for a long time. The true story of Stanley is no doubt just as dramatic but much harder for a layperson to appreciate.
I think the first practical non-military application of autonomous cars will involve a ton of infrastructure. It won't be achieved solely by making the cars as advanced as possible, but by providing a lot of supplemental data from an array of stationary sensors (and processors) installed by a city or theme park that wants to be the first to have autonomous cars.
Eventually human drivers will be banned, and the cars will communicate and cooperate with each other (much better than human drivers!). Traffic engineers will maneuver cars manually in rare instances, and computer-controlled cars will give them a wide berth. Safety will be improved, but so will traffic efficiency. Cars will become less personal, hence smaller and more efficient; crashes will become rarer and safer, hence cars will be smaller; computers will be better drivers, so cars will run faster and closer together. We can look forward to a period of ten to thirty years in which freeways don't get any wider.
Continuing my utopian fantasy, if cars become autonomous and have less personal significance, many city dwellers will choose to use taxi services instead of owning their own cars. That means that most of the cars on the road at a given time can have a sensible capacity, rather than the maximum capacity the owner imagines that he or she might need. Per-capita energy use for personal transportation in the U.S. will drop to a fraction of the current level.
It will happen someday, but maybe not in the next hundred years, depending on how stubborn we are. It would certainly be easier and more rewarding to start with helpful, high-infrastructure environments, but the military has such a massive capacity for funding research that we will probably solve the harder problem of hostile environments first. I.e., we'll have autonomous robot sharks with frickin' laser beams on their heads long before we have Johnny Cab.
They never said that Stanley was using H1 for guidance, just that it was following the same path as H1.
Anyway, congratulations to the Stanford team on a great achievement.
Note to self: get a sig.
"The fact that the Stanford team used a 'tailgating' strategy is the best surprise in the article."
You ruined the suprise for everyone!
Its lasers are constantly teaching its video cameras how to identify drivable terrain, and it knows that it could accelerate more.
Maybe one day it can use its lasers to eliminate obstacles, creating drivable terrain and enabling to accelerate more.
He who knows best knows how little he knows. - Thomas Jefferson
Try the Scientific American article on the DARPA challenge: Innovations from a Robot Rally
It covers all the teams a bit and talks about some of the innovations that were used by the competing teams. It is a little light but worth a minute or your time.
Architectural plans are like computer source code with a couple of differences: You only compile once.
Since 1969, according to the Wiki. I tell no lie here, I can never stop laughing when I realise we already have a Skynet. And it's for our Armed forces.
It was not programmed to tail-gate, it just happened to be so good at what it was doing that it caught up to the CMU vehicle and eventually passed it.
And it was using laser sensors and video cameras to visualize it's enviroment. It's a pretty remarkable system. Makes me wish I'd stayed in school for A.I. programming.
Coding reports in a factory cannot be as much fun as coding a toureg to drive through the desert.
Sean D.
"Hmm. I am to metaphor cheese as metaphor cheese is to transitive verb crackers!"
>What use is a robot that can navigate a desert if it can't actually see anything? It can drive in total darkness, for starters... I'd think that's a big advantage in a warzone. Not sure how rangefinders would cope with sandstorms, mist etc, but then you could maybe switch to another set of non-visual sensors (acoustic? Only when it's 'cam' out there, probably... Why only limit a vehicule's sensors to the visual light spectrum? Our eyes and brains are so good at the task, because it's all we have to orient ourselves, while computers have other alternatives that may prove better suited to a situation. Add the fact that computer-vision is not too good, to put it mildly, in low-light situations, mist, dust etc...
Every deer, cow, buffalo, etc... has a GPS unit strapped to its back.
The desert is essentially a static environment over a short time frame. You can avoid nearly 100 percent of potential accidents by merely arresting your own movement. Compare to a busy highway, where dozens if not hundreds of independently moving actors can impact you regardless of how flawlessly you negotiate the roads.
Planes fly from city to city on autopilot, but jet fighters do not dogfight or land on carriers on autopilot. Same reason--it's a huge jump to go from single actor in static environment to the physical negotiations of multiple independent actors.
Build a man a fire, he's warm for one night. Set him on fire, and he's warm for the rest of his life.
The Elantra is a four door car. There is no reason, what-so-ever, for a baby in a baby-seat to be in the front-seat of that car.
It really helps if you RTFA before you comment. Stanley tailgated for a while, then decided that the hummer was going too slow and passed it. Kinda hard to win a race by tailgating the whole way, no? ;)
Also, the hummer from CMU was using all the GPS data and some data that the team gathered about the route beforehand. Kinda cheating if you ask me. Stan uses a combination of lidar(for under 30 meters) and visual recognizion via camera to extend its viewable range to 80 meters. But on page 4 of the fine article, you can read a more detailed explanation.
Wort Wort Wort!
Still kudos to Stanford for being the fastest against the odds and a fine implementation, but this is hardly a giant leap for AI. Baby steps, useful interesting baby steps.
This sig is intentionally left blank
See page 11 of their DARPA Grand Challenge entry document (http://www.darpa.mil/grandchallenge05/TechPapers/ Stanford.pdf). It runs Linux, the software is written in C/C++, and it uses IPC for communication between the various components.
(This was pointed out by an AC elsewhere in this thread),
try to straddle a squirrel running across the road like I do?
Cheers!
So you are saying Stanley didn't use GPS? I think your link to the article says something different. By the way, on that link, the Odometry is a joke. It is absolutely useless in this application. But since you're an expert in the field, I'm sure you know why that's the case.
Anyway, CMU did spend alot of time and money to collect data but if you read the article you'd see it mentioned that only 2% of the data was at all useful. Imagine if >90% of the data was actually used.
I'm no expert, but it seems to me you're way better off waiting until the squirrel has left the roadway. Even then your thighs must get really scratched up...
My hangup is that with the Robots on this BIG course, why would they be right next to one another? But I guess it's because engineers always think in optimal ways, so the two robots must have been along the same optimal path at the max speed allowed. That's possible, but there's room for doubt.
It must have been exciting to see!
Cheers!As to the scratches, that's really only a problem if you're trying missionary position. If you're going to do this sort of thing, straddling's the right approach, I think.
I read about 5 comments on lane changes and people stepping out in front of it.. lets not forget this is funded by a military research organization, and that is the end goal, not alllowing you to catch some zzz's on your commute to work.
Journal of the trip: NHAA journal and information on the software, RALPH
NHAA showed that it's possible to do at highway speeds (60+ mph), using 1995 technology. The construction issues are a challenge. From the journal, it sounds like RALPH handled construction reasonably well, but there certainly are construction sites that even many humans can't successfully navigate...
If all of the vehicles in your immediate vicinity are traveling at the same speed in the same direction their velocity relative to eachother is 0. You dont have to swerve to avoid a chair across the desk from you, do you? The same will apply to groups of vehicles traveling the highways under computer control.
True as postulated, but nowhere near real world conditions. For starters cars need to accelerate/decelerate along the axis of the road to merge, exit, and find openings to change lanes. Plus they have to accel/decel laterally to accomplish any of those. In addition, there are still going to be points of congestion which will force decel and accel. Computer control doesn't solve the basic problem of too many cars and not enough lanes.
Sociologically it's even more ridiculous...are people really going to be ok with having no control over their speed? Some people are in a hurry and want to go faster. Others like to drive conservatively at low speed. To achieve your assumed conditions there must 100% participation in a controlled traffic system--a classic command-and-control approach. The whole reason the car/road system put the railroads out of business was the personal relevance of full control over your journey.
In addition it would have to be switched over instantaneously, so as not to have computer-controlled and (unpredictable) human-controlled systems sharing the same road. There is no half-way state...if both are on the road the system devolves to the chaotic state of driving today.
Last time I drove down the freeway the only obstacles were other cars.
Not sure where you live, but around here (Mid-Atlantic U.S.) we have traffic jams, slow-moving trucks, random debris (fell off truck, car, etc), broken-down vehicles, various wildlife (up to deer size), puddles, snow patches, ice patches, and occassionally pedestrians on or along our highways at various times. In addition there are occassionally poorly-marked (or unmarked) lane changes around construction areas.
The "corridor" is like the highway and the obstacles simulate other vehicles.
Yet they didn't even allow moving vehicles around each other in the corridor...one had to stop while the other was near. They simply converted any multi-party system to a static system for one of the parties.
I definitely agree that you have to start somewhere and that the Grand Challenge is a great step forward. But let's not trivialize the challenge inherent in driving around other drivers...and in changing sociological expectations.
Build a man a fire, he's warm for one night. Set him on fire, and he's warm for the rest of his life.
"Still kudos to Stanford for being the fastest against the odds and a fine implementation, but this is hardly a giant leap for AI. Baby steps, useful interesting baby steps."
ASIT
" 1. It solves the problem completely.
2. It doesn't require many resources.
3. It doesn't involve negative side effects when used.
4. It is a solution that only a few will find."
You'll find that the standford solution meets the criteria of the above method.
So you are saying Stanley didn't use GPS?
I believe all of the competitors used GPS in some form or another, I was just trying to point out the extensive preprocessing done by humans on the CMU team(and they still lost by 11 minutes). Doesn't seem like an autonomus vehicle should need all that help, kinda defeats the purpose of being autonomus.
But since you're an expert in the field, I'm sure you know why that's the case.
I never claimed to be an expert, I only claimed to have RTFA.
By the way, on that link, the Odometry is a joke. It is absolutely useless in this application.
I agree that the Odometry is a joke on Stanley, especially on sand and other surfaces where the wheel may be spinning, but not going anywhere. I like the solution that a bunch of high school kids came up with for their "Doom Buggy", but I don't see any mention of them finishing the race. They used the same technique that is used for optical mice. I don't know how effective it really was, but it seems like it would be a much better idea than seeing how far the tire rotated over a certain period. That article is here. They talk about the "Doom Buggy" on page 2.
Wort Wort Wort!
Their [CMU's] mission: create a digital map of the race area's topography. The team logged 2,000 miles and built a detailed model of the desolate sagebrush expanses of the Mojave. That was only the beginning. The Red Team purchased high-resolution satellite imagery of the desert and, when Darpa revealed the course on race day, Whittaker had 12 analysts in a tent beside the start line scrutinize the terrain. The analysts identified boulders, fence posts, and ditches so that the two vehicles would not have to wonder whether a fence was a fence. Humans would have already coded it into the map.
Does anyone else think that this makes a mockery of the term "autonomous"? All the route planning (the hardest part of having an autonomous vehicle) is done by humans with the CMU vehicle. It's not autonomous at all.
If I were putting on a competition like this, I wouldn't reveal the location or route beforehand. Instead, I would have all the contestants bring their vehicles to a warehouse, then I would put them in a truck, and drive it to the race start. The race organizers would place them in random order at the starting line, then at 2-hour intervals the officials would press a prominent "Start" button that each vehicle would be required to have.
That would be a true test of autonomy. The current race is clearly biased to favor projects that already receive massive DARPA support.
Global route planning is not the hardest part. This challenge is really about the real-time navigation/driving details. Google Maps can tell me the route to get from San Francisco to New York City. Google Maps cannot drive my car through that route. Now if you want to just throw robots into an unknown course... guess what... the entire time spent by the robots will be building a environment map, getting stuck in deadends and then recovering. Not too exciting especially when mapping can be done much more efficiently and frankly is a solved problem.
I am actually very skeptical that Stanley had no preprocessed map information. Why would a team like Stanford ignore satellite images, elevation datasets? Being able to see 80m ahead is great and all but how do you suppose you can navigate many kilometers without knowing that 500m out there is an uncrossable cliff or a river? The only way you could navigate is by following a road that is already there or by running on a trivial course without obstacles that affect global route planning.
Think about it.
To some 17-year-old who loses 10 cents on every typo he makes (somewhere in an obscure German town), though, this could be a wakeup call for coding more AI into spell-checking. ;-)
I can say, without a doubt, that all teams uses GPS...
The E-Stop boxes are GPS based.
The reason I know this is that I spent a good amount of time in the garages at the NQE (except from noon to one when I escaped the heat by going the the Internet center to read slashdot), and saw the announcements that "Teams are to keep their E-Stop recievers off in the garage area because, WITHOUT A GPS FIX, they cause interference.
Then let the robots download maps, etc. autonomously (automatically), and as needed, during the race.
And if route planning is so easy, why did CMU need "12 analysts in a tent beside the start line scrutinize the terrain" which "identified boulders, fence posts, and ditches so that the two vehicles would not have to wonder whether a fence was a fence".
That's not "autonomy", it's glorified R/C.
The DOD, major corporations, government, etc. are all concerned with limiting their liablility by limiting what a vehicle can do using AI technology - not about improving the performance of the vehicle. There is an effort to create a truck that can drive itself across the desert at 30 mph, but not an effort to make one that can drive across the desert at 150mph by integrating more seemlessly with the human driver.
Don't get me wrong. I would love to have a car that could drive me to work each day while I read a book or browsed the net. But the current push in vehicle control systems is more toward limiting the performance the driver can expect than improving it. When I stomp down on the accelerator I need to go fast, right then, not have my acceleration limited by the fuel injection control system - and with he cost of gasoline as high as it, I am not going to stomp on the accelerator unless I have a good reason.
So much of our intellectual efforts, engineering and legal expertise is directed toward limiting liability and risk aversion that it will eventually cripple us.
If we lose our will to push the boundaries of what is capable, then we are doomed.
Here's a PDF on probabilistic robotics.
d f
http://robots.stanford.edu/papers/thrun.probrob.p
http://robots.stanford.edu/papers.html
Has more references.
The quote you posted proves exactly my point. Your first post stated that "all the route planning is done by humans". You said this was the hardest part. I responded that route planning is easy. I said that real-time navigation/driving is difficult. Now you post that these 12 analysts were not doing route planning. They were trying to help the navigation/driving sensors systems by annotating features so that the vision and laser sensors would not "wonder whether a fence was a fence".
You also happened to entirely skip responding to my point that only the most trivial of race courses would not require map information for global route planning.
Your notion of "autonomous" vehicle operation needs more elaboration. How autonomous can any of these vehicles be? What happens when they run out of gas? Can they fill up their own tanks? If not, then what happens when the race spans 500 miles instead of 132? Autonomy in robotics at the most basic level is the ability for a robot to act based on sensor inputs without direct human invervention. All of these vehicles pass that test with flying colors. Now you may think that's not autonomous enough. Fine. Give me your precise definition of autonomous. Please take into account my example of the vehicle running out of gas during the race.
If you read the red team website you will find that H1lander was supposed to go faster than stanley but did not due to engine problems. Now that I know that the red team used a classical planning approach while stanford used a real adaptive approach I think it's good that stanley came out on top.
I wonder by the way if the H1lander was planning on breaking the speed limit. Thurn said they did during qualification (look at the qualification video's on staleys website, they are great!)