Another Step Towards the Driverless Car 224
jtogel writes "At Essex, we have for some time been working on automatically learning how to race cars in simulation. It turns out that a combination of evolutionary algorithms and neural networks can learn how to beat all humans in racing games, and also come up with some quite interesting, novel behaviours, which might one day make their way into commercial racing games. While this is simulation, the race is now on for the real thing — we are setting up a competition for AI developers, where the goal is to win a race between model cars on real tracks. As the cars will be around half a meter long, the cost of participating will be a fraction of that for the famous DARPA Grand Challenge, whereas the challenges will be similar in terms of computer vision and AI."
In case of rapture (Score:5, Insightful)
Seriously, this is a technology whose time has come. Persuading elderly drivers to give up their cars is difficult, and the baby boom generation is putting a lot of people in that situation in the next decade or two.
Re:In case of rapture (Score:4, Insightful)
I still would have to be VERY VERY sure of the system and see it tested out the wazoo before I would ever consider getting in one. I don't trust a team of EE and SE specialists to think of all the possible reactions they would need coded in for outlier situations. Even an autopilot for an airplane doesnt have to worry about falling trees, landslides, or elk...unless its a REALLY REALLY bad day at least...
Safety vs. Freedom , again. (Score:5, Interesting)
Now this, to me, is a very important distiction. What if for this to work well, all the cars have to be computer controlled? What if computer control is then mandated? This is a whole new exciting level of "nanny government". Sure this might be safer in that there would be fewer auto accidents, but do you really want all transportation to be centrally controlled? Sure each car might be autonomous at first, but emergency workers need the ability to remotely turn on off, right? It's for everyone's safety.
Re:Safety vs. Freedom , again. (Score:5, Insightful)
But on the other hand, getting killed in an automobile is much too common, especially given that almost everyone has to travel in one at some point, if not very frequently. Getting around shouldn't be so bloody dangerous considering how ubiquitous it is. Imagine not every having to let drunks choose between being responsible vs driving home drunk. And imagine not ever having to be on the road where some random drunk or incompetent driver, can end your life at any instant, where it is just bad luck that puts you in this spot.
Automobiles are an outdated and obsolete technology, or at least should be. The problem is coming up with and implementing the "next step" when the current technology is so ingrained into our society and city planning. It is a very non-trivial problem to come up with something better, and another non-trivial problem to "upgrade" to that something better on a live production world.
Re: (Score:2)
The other posters fears about unaccounted for circumstances are unfounded. If there was some problem with the program it would default to safe mode (eg car would st
Re: (Score:3, Interesting)
That would be great if there was a way to make sure every other car would stop too.
There are some differences
Re:Safety vs. Freedom , again. (Score:5, Insightful)
The safety aspect is definitely a selling point. But that's not the killer app.
The killer app for AI cars is: traffic throughput. Right now, traffic throughput is limited by our need to leave lots of space in front of our car so that we don't hit the guy in front of us. This creates low throughput through traffic lights because everyone must wait for the person in front of them to move away, before starting to move too. AI would need none of that.
Ditto for freeway merges and weaves. AI could weave two lanes of cars together with ease... even without central automation or inter-car communication. All that is needed is a sufficiently standardized algorithm or a sufficiently clever computer. Our brains already do the same, even when the "DriveCar.exe" process is set to low priority in favor of a ringing cellphone.
Can you imagine how fast traffic could move if (to name just one benefit) everyone rolled forward instantly when the light turned green? And if nobody slowed down to rubberneck a roadside accident?
The implementation problem might solve itself too. Once AI cars prove their mettle, I can imagine that cities will designate more and more lanes as "AI only", with attendant increases in speed limit and throughput. Sort of like how HOV lanes work today. Soon we'll all be clamoring for an AI car in order to get the same benefits.
Re:Safety vs. Freedom , again. (Score:5, Funny)
Ehm. no it isn't (Score:2)
The killer app for AI cars is: traffic throughput. Right now, traffic throughput is limited by our need to leave lots of space in front of our car so that we don't hit the guy in front of us.
Um, no it aint.
Traffic throughput is limited by how quickly you can get cars off a particular road. i.e. Parking. You can have cars 1/10th of a second apart, but if it takes 10 seconds to park the thing, or even 5 seconds to turn onto another road, that is the limiting factor.
If everyone has their own car, road performance is limited by parking bandwidth. Now, if everyone used a taxi...
Re: (Score:2)
In some situations, this might actually be the solution. You don't own a vehicle, you have taxi service with an autonomous taxi. It drops you off at work, and then drives on to its next pickup. Smart algorithms mean it moves on to the nearest person with autonomous taxi needs. Smarter algorithms might allow the taxi to share taxi service, perhaps with isolated compartments for each person.
You
Re: (Score:2)
You run into the problem, though, of commuter patterns, as you have a large group of people moving from the 'burbs to the city in the morning, and then all moving back in the afternoon.
You just need a cache of taxis at the burb/city end large enough to keep the service time down to within acceptable limits, say a 120 second wait. Bearing in mind that the taxis re-circulate once dropping off their passengers the cache doesn't have to be anything like as large as you might think.
...
And it can be done today without requiring A.I.
http://faculty.washington.edu/jbs/itrans/prtquick. htm [washington.edu]
http://www.atsltd.co.uk/ [atsltd.co.uk]
http://www.personalrapidtransit.com/ [personalrapidtransit.com]
The performance of a PRT system, like roads, is d
Re:Safety vs. Freedom , again. (Score:4, Insightful)
Yes, and later, traffic throughput will be limited by the need to leave lots of space in front of the car so that if the car in front of us suffers an equipment failure (not necessarily computer-related; a blowout qualifies) the computers have time to work out a solution that doesn't involve collision.
Certain realities of physics make it a good idea (or even a necessity) to not have cars tailgating one another, even if they are automated and much better drivers than humans are.
It would make far more sense just to replace the highways with rail lines, load cars onto trains, and ship them places. Then the cars can be loaded literally on top of one another.
Re: (Score:2)
Re: (Score:2)
Not to mention the fact that half of the people on the planet do not live in cities, and these are the people who usually have to travel farthest. City planning does not universally solve the problem of transportation.
Re: (Score:2)
Actually the chances of getting killed in a automobile accident are pretty low compared to other forms of death. I'm not sure I will trade in my freedom for the security of these low changes. Simple things like following the posted speed limit and slowing down when condition deteriorate and looking at the other traffic and hazards while your driving can change
Re: (Score:2)
Re: (Score:3, Informative)
The page you listed showed figures of 43,354 people dies form injuries sustained ion a car crash. I have figure for 2003, or 2004. Assuming they are the same (and I don't believe the war deaths are counted because they are overseas and not in the US.) or similar in 2004, the totale number of deaths in the US was 2,398,343. This means that 2,354,989 Died from other causes. Thats a pretty big number compared
Re:Safety vs. Freedom , again. (Score:5, Insightful)
Re: (Score:2)
I'm not trying to take anything away or belittle their deaths.
Re: (Score:2)
The OP was comparing the # of auto-related deaths to the *total* # of deaths. Death is inevitable, obviously, but there are ways to put it off for a while.
Re: (Score:2)
Well, that's true. Twice this many people die every year just from bad drug interactions or drug allergies.
You're actually more likely to be killed by the hospital than just abou
Re: (Score:2)
Re:Safety vs. Freedom , again. (Score:4, Funny)
wow, what country are you writing from? (Score:3, Interesting)
Re:In case of rapture (Score:4, Insightful)
You're probably young, so the ageism can be explained.
I know people in their 80s, perfectly capable of driving, and renewing their license.
We don't need dorks mandating new technologies to use. We just need the DMV to do it's job, which is to make sure only qualified drivers are qualified to drive.
Re:In case of rapture (Score:5, Insightful)
It isn't the middle-aged who oppose this plan.
testing & riding; was: re: In Case of Rapture (Score:2)
If the issue is the Rapture, I think most of the people left behind will have tougher aspects of their lives than whether their cars will function safely.
As far as testing goes, when we were working on the jeep two years ago, we had a nice, big, back yard to play, courtesy of as he [www.scottajones] has considerable acreage (20? 30? 40?). It cuts down additional neighbors and there's a lot of nature preserve. You can't hear the traffic a few blocks east of it from the major thoroughfare coming it's like a nature preserve
Re: (Score:2)
Something which most people don't think about is: if the front end crumples too much, the engine has to go somewhere. It certainly doesn't jump out from under the hood and clear itself from the car. It's not going to deflate into a small piece of rubber.
Nor does it necessarily have to move straight back into the passenger compartment. BMW has made it a selling point to illustrate how their cars are designed such that the forward drive train (engine + transmission) breaks free of the body and is deflected down, so that the passenger compartment rides up over the top of it. Many other high end luxury cars are similarly designed.
Re: (Score:2)
Practically all cars are designed to do this. The engine and transmission mounts break (nearly all of them consist of two metal plates connected to one another only by a piece of rubber vulcanized between them) and
Re: (Score:2)
Safety. Yet another compelling reason to drive a mid-engined car [wikipedia.org]
Re: (Score:2)
Re: (Score:2)
You also have to consider how resilient such a system could be to tampering. If it relies on GPS (and I don't see how it wouldn't), then what will it do if I jam the GPS signal? Will the car know how to get off the road safely with the GPS data jammed? Can it still get to safety if I jam the signal near a busy highway and there are hundreds of cars trying to pull over at the same time?
Don't be daft. You're overthinking it. It certainly wouldn't use GPS for anything more than coarse navigation, as even an optimistic 2 meter accuracy is insufficient to keep you in your own lane. Maneuver will likely be managed with optical sensors. Is there a problem with people shining spotlights in each others eyes on the road today? Unlikely it will be a problem when it's robots driving either.
Re: (Score:2)
Re: (Score:2)
Re:In case of rapture (Score:5, Funny)
Have you ever really tried? Sure they get ornery and wave their canes around a bit, but most of them are fairly frail and the task can be completed with ease. Sometimes they're confused and just think you're a valet - these ones will hand you the keys with a smile!
Re:In case of rapture or old folks driving (Score:5, Funny)
I find backing over them works fairly well.
Not just for older drivers (Score:3, Interesting)
Lower insurance premiums - and if the car has an fender bender, I can point to the manufacturer and hopefully won't be branded as an unsafe driver for life if I didn't do th
Re: (Score:3, Insightful)
Why not catch a train or bus to work?
Re: (Score:3, Insightful)
Bus: Have a stop locally - 5 minutes walk. Take bus to a central station - 1 hour 5 minutes. Take second bus, unknown wait, to destination - 40 more minutes. 10 minutes walk to destination. 1h55m minimum.
No thanks.
Train? None here. I don't live in the city.
Re: (Score:2)
Not just for older drivers - wives also! (Score:5, Funny)
Sitting in a car with my missus driving is much the same as being in a driverless car:
Biggest difference is that the thing is more likely to know the way to someplace.
Re: (Score:2)
Re: (Score:2)
Just tell them how easy it will be to find the farmers market...
Re:In case of rapture (Score:5, Insightful)
Not just, consider accident caused by drunk drivers, by drivers fell asleep, careless drivers...
But don't expect a smooth transition. An "AI" driver could silently save thousands of lives, but the first cases where the AI was the reason for an accident will cause major outcries.
It's the nature of human beings to react like that.
Re: (Score:2)
I do think we'll see unmanned transport planes first, though. Flying is easy. There's less to crash into, you only land at a pre-set number of known locations, and planes can be much more expensive than cars to cover the costs of new technology.
Re: (Score:2)
I do think we'll see unmanned transport planes first, though. Flying is easy. There's less to crash into, you only land at a pre-set number of known locations, and planes can be much more expensive than cars to cover the costs of new technology.
That's insightful, although I wonder if they will want to shift so much responsibilities for a so expensive machine onto AI for a plane. A pilot
Where are the Competition Specifications? (Score:2, Interesting)
The competition sounds like a manageable project for academics (versus the DARPA event).
Is the competition still in the vapor-ware or maybe-someday stage?
Anyone have a link (perhaps IEEE) that has details?
Re: (Score:2)
The issue of DARPA is still to be decided. There's a Grand Challenge 3 this fall, two years after #2 completed the desert challenge; i.e., they figured people would need two years for preparation. And that is for realistic traffic. Six hours, six miles...in Europe. There's a demo to be met by April 13. I'd be comfortable saying I've received 200 messages so far. They've been coming in pretty regularly since the May 1, '6 News Release (a two page PDF).
Re: (Score:2)
But urban settings with a lot of pedestrian traffic are a complete technical and legal nightmare. Then there's the issue of very poor quality roads or severe weather
Re: (Score:2)
Small aircraft don't. Not unless you're very lucky anyway!
Re: (Score:3, Funny)
Re: (Score:2)
Re: (Score:3, Insightful)
Drunk driving, Sleep driving, Stupid driving, Sick driving...
Seriously, most people would benefit from an RELIABLE A.I. system to drive for them.
Once AI systems are more reliable them us, then the time will be here.
Re: (Score:2, Informative)
Oblig (Score:2, Funny)
Forza 2 (Score:2)
Re:Forza 2 (Score:5, Interesting)
We have ourselves gotten player modelling working fine with evolutionary neural networks, which can generalize, but the Forza team didn't consider these techniques reliable and fast enough in time for the release of the original game. Maybe things have changed with Forza 2.
There is some information on the Forza AI on http://research.microsoft.com/mlp/forza/ [microsoft.com], and our approach to modelling is described in http://julian.togelius.com/Togelius2006Making.pdf [togelius.com]
Note that all this is about modelling behaviour, not about creating new behaviour from scratch; there are some papers on this on my website [togelius.com] as well.
Reminds me of RARS (Score:2)
Anyone have first-hand experience of how that is coming along? Can you program these with neural nets also, or still "just" hand-coded? (My own experience is nigh on a decade old.)
Surpassed by TORCS (Score:2, Informative)
There have been several robots that use various learning techniques, though none to my knowledge have been full-blown AI/neural net solutions. To be honest, I query the advantages of doing it that way. A robot that has code to plan a smooth & optimal path around the track & calculates braking and steering accordingly will do much better (initially at lea
say goodknight (Score:2)
Am I the only one? (Score:2)
Traffic (Score:3, Interesting)
Re: (Score:2)
Re: (Score:3, Interesting)
Re: (Score:2)
The underlying research (Score:5, Informative)
Some of them are of course better than others. I can recommend this one [togelius.com], about evolving general and specific driving skills, this one [togelius.com] about co-evolution, this one [togelius.com] about different learning techniques, and this one [togelius.com] about modelling human driving and evolving tracks. There are several new ones, including one on physical cars, which are not on the website yet - mail me if you want a preprint!
All this assuming that anyone actually reads academic papers... sometimes it seems that not even the guy who writes the paper actually reads it. (Not true in my case, of course!)
Re: (Score:2)
The paper seems very game focused and the extrapolation to real driving seems a bit distant, but I concede not reading the other papers.
One comment I have about your track modelling: all your track
Never gonna happen... (Score:2, Funny)
The problem isn't in the technology (Score:3, Interesting)
Many manufacturers have been dissuaded from pursuing the technology and installing in their vehicles because in the case of any accident the corporation would be liable. Obviously the 'driver' wouldn't be at fault because they wouldn't be driving.
No large corporation is going to put itself in line to pay out on every bump, scrape and minor slaying caused when their killer robo-cars Attack!
Re: (Score:2)
Re: (Score:3, Insightful)
Power steering? What if the pump fails during a hard turn?
Air bags? What if some idiot doesn't buckle up and gets killed by the bag?
Seat belts? What if the latch weakens with age and breaks during the impact?
Windshield wipers? What if they crap out in the middle of a heavy rainstorm?!
Sooner or later the technology becomes mature enough that the benefits outweigh the risk of liability. At that point, the manufacturer slaps another disclaimer in
Oh great! (Score:5, Funny)
The environmentalists will not be happy with this development!
And try this against a real racer (Score:2, Insightful)
Implying that these cars could "drive themselves" in any meaningful or safe
Re: (Score:2)
Re: (Score:2)
Re: (Score:2, Interesting)
A boringly smooth road track with an entirely predictable car, like most F1 tracks and cars are becoming, would be the ideal combination for these algorithms. No one passes anyway, so maintaining position until a perfect pit stop occurs is the way to get ahead. Good pit
options (Score:2)
Re: (Score:2)
But if you give these cars the ability to communicate with each other and "smart roads", you might be able to end up with behaviors that are just not possible with current cars.
Given your scenario, it maybe possible the car can slam to a complete stop - while communicatin
difficult to back up (Score:2)
Better Yet (Score:2, Informative)
I'm all in favor of robot contests and all but more important, from my point of view, is the ability to share resources (such as test environments, robot chassis, sensors, vision-processing code, etc.) outside of the competition itself.
The biggest unnecessary impediment to robotic research right now, as I see it, is the difficulty researchers have in making comparisons between systems. You demonstrate your racing code on your robot in your test environment. I demo my code on my bot on my test track. Th
Re: (Score:2)
* The simulator should be 100% cross platform (I regulary use Ubuntu, Mac and Windows, and researchers who want to participate in the competition can well b
One big traffic jam (Score:2)
Only works on wide roads (Score:4, Informative)
Notice that in all the examples, the road is much wider than the car. That's not by accident.
Driving using reactive behaviors is easy if you have plenty of room. On narrow roads, though, those approaches fail. You have to look ahead. In fact, to drive in the real world, you need a controller that plots at least an S-curve ahead. Otherwise, you'll end up in a tight spot pointed in a direction that won't get you through.
You don't necessarily have to "plan", in the AI sense, but you need a fairly good dynamics prediction capability, after which you can run a reactive controller on the prediction.
We went through this with our DARPA Grand Challenge vehicle. We started out with a reactive planner, but it just couldn't deal with tight spots. Most of the other teams ended up with S-curve planners, too. The reason you need S-curves is that you need to be able to achieve both a desired position and direction at a point ahead of the vehicle. So you need a curve with at least two degrees of freedom.
The predictor needs to know enough about the vehicle dynamics to make reasonable predictions. For example, predicted S-curves have to be built knowing how fast you can change the steering angle and how tightly you can turn given the current speed and ground bank.
If you need to do this stuff, read up on adaptive model-based feedforward control. The idea is that you have a system that learns how the system behaves as the inputs change and builds a model. Inverting the model gives you a predictor. Given a predictor, you can control.
A useful feature of that approach is that, while you're using one predictor, you can be training a better one safely. Predictors are trained by watching; they don't have to be in control. So you can start out with some dumb controller and work your way up to better ones, without crashing. This is probably how mammals learn motor skills.
Re:Only works on wide roads (Score:4, Informative)
In the case of our (admittedly simple) model, we have a limited line of sight, and I think a good reactive controller can perform optimally (however you define that, often optimality is just another buzzword) given the limited sensor data. We did try evolving reactive control on much narrower tracks with good results - see for example our papers on track evolution. What the controller learns is often just to slow down when coming up to a narrow passage.
Re: (Score:2)
I read the paper and made a copy of the most difficult map, track #8. I'll try to convert that into a form the Overbot software can handle, and run it.
That map, assuming the vehicle is 2m wide and 3m long, has a narrowest width of 4 meters, which is not all that tight. The simulated vehicles in the article seem to have a rather tight minimum turning radius. They do skid in turns; unclear if the controller is exploiting that behavior to tighten turns. What is the minimum turning radius of those vehicle
Re: (Score:2)
Drifting! (Score:2)
Just a little future sight here (Score:2, Insightful)
Some posters have posited the "what if every car were an AI car" scenario. But the transformations all talk about changing our current cars into AI cars.
But I'd like to propose a different slant. If every car is an AI car, then how will this differ significantly from a distributed form of public transportation? If you break it down, the daily commute is filled with SUVs and a lone driver, with the SUV remaining parked (taking up space) for the whole day. So what about "transportion as a service" here (A
A great step forward (Score:2)
Driving in the wrong direction (Score:5, Insightful)
Why are we trying so hard to make something designed to be operated by a human computerized so it stays on the road when we can make a road with rails on it?
Towards (Score:2)
--Richards
Frederik Pohl (Score:2)
Sigh.
Re: (Score:2)
Re: (Score:2)
Of course, that isn't necessarily a bad thing.
Re: (Score:2)
Re: (Score:3, Funny)
Re: (Score:2)
Stuck against some wall is Microsoft market-speak for "crashed and burned."
Re: (Score:2)
For years I've been thinking that we should make something like a TCP/IP for traffic (taking into account limited acceleration and taking collision avoidance much more seriously of course). Self-steering and indeed, self-driving, cars should make this a whole lot more plausible.
The difficulty in a change-over will as always lie in dealing with the part of the population who does not want t
Re: (Score:2)
Re: (Score:2)