Preparing for the DARPA Autonomous Vehicle Challenge

Little Hamster writes "Post-gazette.com has an interesting article on the DARPA funded 200-mile autonomous vehicle race across the California-Nevada desert. They interviewed teams from two of the early favourites, Carnegie Mellon University and the California Institute of Technology. The teams talked about challenges on driving at high speed over a combination of roads, rough terrain and brush-covered desert, where the robot would need to consider how fast it can make a turn, the possibility of spinning tires and the potential to become airborne when hitting bumps."

Comments

Now we're onto something...

[Anonymous Coward]

Just mount a versalaser on it, remove the laser safety shroud, set it loose in Iraq and let it carve "WWJD" on all the terrorist's foreheads...

They'd kill each other after that...

Predicting future positions of the vehicle in RT

[Anonymous Coward]

And if the source code that runs the thing ever goes public, then the vehicle gets remarkably easier to target, given another computer and a detailed topomap of the area.

Further applications

[digital bath]

This is really cool.. technology like this could be used in consumer cars to reduce rollover/tire spin/etc. Maybe even 'smart' cars that drive themselves, leaving the human passengers free to sleep or get work done.

Re:Further applications

[DogIsMyCoprocessor]

leaving the human passengers free to sleep or get work done.

A nice thought, but in reality the passengers would just piss the time away reading Slashdot over a cell phone connection.

Re:Further applications

[kfg]

Traction control is old hat. On an electric vehicle which uses a seperate motor to drive each wheel (the ideal setup, eliminating most of the drive train and making a differential totally redundant)it's actually not even that difficult. A few sensors and some computer logic to moderate the electricity to each motor, providing a specified amount of torque at each corner.

Pretty sweet setup really. Too bad about the whole battery thingy.

What might surprise you though is that Ford is already not only already testing totally autonomous vehicles, but they're at such an advanced state they're doing so on public roads in heavy traffic. So far they're doing just fine.

It's kind of odd to watch someone sitting in the "driver's" seat, and not driving.

And no upgrade to infrastructure needed, as was posited in the "old days" (guide wires under the roads and such). Advances in computer, feedback and sensor technology have made all that sort of thingy obsolete.

So maybe in the future your fridge will not only call up your grocer when you're out of milk, your car will go get it for you while you sleep.

I'm not sure about being able to train my cat to go out to the car and bring it in though. I'll have to trade her in on an Aibo or something.

KFG

Re:Further applications

[Osty]

This is really cool.. technology like this could be used in consumer cars to reduce rollover/tire spin/etc. Maybe even 'smart' cars that drive themselves, leaving the human passengers free to sleep or get work done.

We already have all of that technology available already.

• Preventing roll-overs: Buy a car that's not top-heavy. If you have a real need for an SUV that is top-heavy, don't try to drive it like a car, because it's not. It's a truck, and you should be aware of that (ie, avoid turning sharply, braking suddenly, etc). The newer cross-over and car-based SUVs (Chrysler Pacifica, Porsche Cayenne/VW Touareg, Infiniti FX models, etc) are much better in this respect. I'm referring mostly to the body-on-frame truck-based SUVs. I don't drive my huge F250 like I do my Boxster, simply because the F250 doesn't handle like the Boxster does.
• Wheel spin: Traction control/stability management systems are quite intelligent these days, using modulated application of braking at the different wheels depending on what's need. Otherwise, keep your tires in good condition (check your remaining tread depth, air pressure, etc) and use the proper type of tire (summer ultra-performance tires are dangerous on snow or ice, of course) and you'll be much safer. It's scary the number of cars I see on the road with bald tires or low pressure.
• Cars that drive themselves: Busses, trains, etc. Of course, this assumes you're in an area with a good mass transit system, which many of us are not. On the whole, though, I'd rather entrust myself to a human bus driver than an autonomous car, at least for the forseeable future. (That said, I never use mass transit, because it's simply not useful where I live, and I love driving :)

Okay, so those may not be as glamorous as a fully-robotic car, but the technology is already there. And as far as future autonomous cars go, so long as I can still buy a car that lets me manage throttle, brakes, shifting on my own for fun, I'll be happy.

Re:Who Do You Want to Shoot Today?

[digital bath]

But some of it ends up benefiting the masses as well - like that little thing called the 'internet' that you used to post that comment detracting DARPA, for one ;)

Sounds dangerous

[El]

Isn't anybody concerned about autonoumous vehicles running over spectators? "Hey, I thought it was a rock." I certainly wouldn't trust my life to a DARPA visual recognition system.

Screw flying cars

[Michael Crutcher]

.. what I really want is for everyone to have self driving cars. It's really much more efficient to roll a weight across a smooth path than it is to fly it over the same path. I'd like to be laying back on my couch in my car going 200.

That would be cool.

Re:Screw flying cars

[toasted_calamari]

Microsoft Autodrive XP has performed an illegal operation and will be shut down. Please stand by while Microsoft Autodrive XP sends an error report in order to serve you bet(&*....[CARRIER LOST]

Robots entertaining robots

[apoplectic]

Get these robots to drive NASCAR to entertain the automatons known as fans.

Just like a video game

[BagOBones]

You ever wonder if video game developers haven't already solved many of the AI issues in terms of driving?

The real problem would be getting the track information in real time and telling the AI what each object is.

Then again I am sure game developers get to cut a bunch of corners because the AI knows the track perfectly because it is a part of its system.

Re:Just like a video game

[mrseigen]

There are probably lots of game hackers trying to use the best of the best in pathfinding algorithms. It'll be interesting to see if what the game industry has been using for years actually does keep a speeding vehicle from getting embedded in rocks, children, and sandstone walls.

Re:Just like a video game

[nissin]

Yes and no. As you say, you need to get the AI information about each object and what it is. I'm working on the Caltech team, and that is where the majority of time is being spent on the software side - detecting and classifying objects. It is an extremely difficult task given todays technology, a limited budget, and the variety of obstacles you can find. Some examples:

How do you determine if there is an object, or it is just mud on the camera?
How do you detect dust and filter that out?
How do you detect a fence - the links are generally too small to be picked up on the camera until you are very close.
How do you detect water?

The list goes on and on...Some of these have answers, some do not. Many times you can use a variety of sensors - visual, ladar, inertial, gps, etc and at least one of them will be giving accurate information. But how do you deal with inconsistent information? GPS says you are in the middle of a river because it is off by a few meters, but the visual says you are not.

I would say that once you have that information, however, the problem becomes relatively straight forward.

Re:Just like a video game

[Effugas]

How do you determine if there is an object, or it is just mud on the camera?

Short answer: Saccades.
Longer answer: Put a motor on your camera that allows limited angular motion. Put teflon-coated plexiglass in front of the camera (probably with a wiper). Saccade. Check angular parallax.
Alternative Answer: Dual cameras.

How do you detect dust and filter that out?

Do horrible things to shadow elimination code.

How do you detect a fence - the links are generally too small to be picked up on the camera until you are very close.

See if you can find a digital camera vendor that will let you hack their firmware. Take 5mpix scans, have the camera output regions with consistent high frequency high contrast shifts.

Much better is to do what people do, and look for the posts between fences.

I don't think reflectance will be helpful for this.

How do you detect water?

I do think reflectance will be helpful for this.

A bigger problem is that people are trying to do way too much with GPS...GPS ultimately says what _was_. Using GPS to avoid driving into a lake is a good way to drown.

Is there a mandatory minimum size for the cars?

--Dan

Something to taxi the average cellphone user

[Coolmoe]

About time that somebody did something about dangerous cellphone users. I would use one after the technology gets a good track record. I could work on a laptop and still get to work. Nice for all the drunks out there after work too. Stop by the bar and still get safely home.

A Real Change

[cluckshot]

The first posters here don't have a clue as to the effects and circumstances of this. The purpose is not Autonomous Kill Vehicles though it might occur. Cruise Missiles etc already do this as does the Predator to one degree or another. The purpose here is to reduce the overhead cost on the army dramatically in hauling supplies etc over long distances with or without roads. To do this you need vehicles than can bypass disabled vehicles and overcome obstacles. They need to be free of drivers who get tired and eat up supplies.

The real effect here will be civilian. The project which like it or not will happen regardless of DARPA someday soon, is going to very nearly completely alter how we live.

To illustrate: suppose you are old blind and unable to drive. (It happens to the best of us) Now you will be able to go where you want without somebody driving you. Suppose you want to go to work but don't want to own a car? Mass Transit? No! you just get on your cell phone and call for a car. It arrives shortly and takes you where you want to go and without a driver. Freight? No more Truck Drivers and the wreaks from them being too tired. No more Taxi Drivers. Close most of the Hospitals because wreaks are not filling them up. Kids will not need parents to drive them somewhere.

There is very nearly nothing more profound than this race! It will reorganize our world. The issue here is how will we adapt. This isn't an esoteric question. We had better face it now.

For the Luddites amung us, give it up. Stopping DARPA will only give the technological edge to China. They will do the work. This is a very high amplification Technology. It Amplifies People a LOT. The issue as always will be the morals of those being Amplified, and will we allow this to cause others to be lost in the "noise."

Re:A Real Change

[khenson]

You know... You're right. And reading your article I was getting more and more riled up about removing YET another right to the few things I actually enjoy on this planet - driving.

But the more I think about it the more I like the idea of having alternative transportation for those deemed "incapable" of driving a vehicle.

This way you would have to earn the right to actually control the vehicle you drive. We could test drivers like other countries do - inclement weather condition tests, obstacle tests, reaction tests, vehicle control skill measurements, etc.

I have raced professionally (mostly dirt track) and I can tell you that many people who have licenses shouldn't be allowed to drive a vacuum cleaner.

This actually make a wonderful solution...

Re:Who Do You Want to Shoot Today?

[cK-Gunslinger]

but 90% of it's about killing people, making nuclear war more practical, or causing other kinds of evil and terrorism,

Oh come on! What kind of crap is this? Perhaps you've heard of that little DARPA creation called the INTARWEB?! =P

No one wants to make nuclear war 'practical.' 90% of research is about NOT killing people, as killing a lot of people typically doesn't help win wars. This isn't the middle ages where you can hope to wipe out an entire society in a single war. What DARPA is interested in is destroying *targets* - things like launchers, tanks, fighters, satellite links, etc. Successful live tests are those that *minimize* casualties, not maximize them. We've had the technology to maximize death for decades now.

EXPERIMENT: Please do not mod down

[ThisIsAnExampleAccou]

ThisIsAnExampleAccount@yahoo.com
Please do not mod this message down. I am currently running an experiment to see which spambots are collecting addresses from Slashdot. Please do not mod down. In fact, if you could mod this up, that would be great. I will publish the results of this experiment as soon as it is complete. hopefully the results will be usefull to all /. denizens. Thanks! ThisIsAnExampleAccount@yahoo.com

Me too, me too.

[spoco2]

I would also like to test an e-mail address here too bill.gates@microsoft.com, just want to see how much SPAM it gets... just a test people...

:)

Desert... pshaw...

[ssclift]

Let's see them drive that beast through Bruce County Ontario blizzard...

.... tires rumble left ... time to steer right a bit ....

.... tires rumble right ... time to steer left a bit ....

.... honey, can you open the door and tell me how far I am from the edge of the road ....

Re:Wrong

[cyril3]

Or perhaps you'd rather we just threw the money into some secret black box projects, never to see the light of day?

Well the Blackbird was pretty cool. In fact it was the coolest damm piece of tech so far developed.

And I for one would rather see them flying around taking pictures than a bunch of autonomous laser tanks trying to miss civilians as they take out the eye of some dumb third world conscripted grunt who happens to be wearing the uniform of the 'enemy de jour' just so joe sixpack can read the paper while he 'drives' to work in his SUV.

Q: What do you call a motorized transport wherein you can freely read the newspaper and converse with fellow travellers and not need to worry about passing traffic?

A: The Bus.

Join Team Overbot - no pay, some risk, big prize

[Animats]

That's one of the best articles I've seen on this event.

I run one of the Grand Challenge teams, Team Overbot. We have a vehicle (a modified six wheel drive Polaris Ranger), a shop in Redwood City, funding, equipment, and people. We're well along; the vehicle has most of its actuators and some of the sensors working, and about a third of the software is running. We're one of the five DARPA-accepted teams.

Many of us are Stanford alumni or students, but this is not a Stanford project.

• We need three more good programmers in Silicon Valley. "No pay, some risk, a fraction of the prize." If you're interested, we want to see 1000 lines of C++ you're proud of. You'll need to put in at least 250 hours between now and March. Click here to join.

Our basic technical approach is to build a rugged, reliable vehicle with conservative control strategies. Others may be faster, but we expect they'll get into trouble at high speed. Our top speed is 40MPH. The real problem with the Grand Challenge is not going fast on the easy parts; it's getting through the hard parts.

The 6WD chassis we're using is one of the most bump-tolerant platforms around. It can go over railroad ties at top speed without problems and without going airborne. The center of gravity is low. The front and mid axles have independent suspension; the rear axle is a swing arm. This simplifies low-level vehicle control. All wheels can be driven, although at higher speeds, we will switch from 6WD to 4WD.

We have five computers on board. Three are small PC/104 machines, and two are Pentium 4 machines. All run QNX (the OS for when it has to work.) All are industrial-strength ruggedized units. The actuators are all servomotors driven by industrial microcontrollers. All this hardware is off-the-shelf industrial control gear.

Sensors include LIDAR, doppler RADAR, sonars, cameras, INS, GPS, etc. Some of them are used in unusual ways. That's all I'll say about that.

The pathfinding strategy is indeed borrowed from video game technology. It's more structured than Brooks-type behavior based robotics, and it's less structured than Latoumbe-type planning. There are three layers of control; the top one we call the "back seat driver", because it has only advisory authority over the "driver".

We have road map and topo data onboard, but it's used more as a hint than as rigid guidance. We take the waypoints DARPA gives us (on a CD, at 0430 hrs the morning of the race) and load it in. There's no offline preplanning. Wouldn't help in the real world.

If nobody wins this year, which is quite likely, we'll be back next year with a faster vehicle.

Post questions and I'll answer them here.

John Nagle
Team Overbot

Deterministic agents

[autopr0n]

Well, it will always be "easy" for a computer to figure out the 'optimal' path through terrain. (of course, by "easy" I mean "NP-hard", but a computer could figure out the best easily-findable paths that another computer would probably use)

This gets into game theory, i.e. choosing certain sub-optimal paths in order to reduce 'predictability'.

If you're using simulated annealing or genetic algorithms to find a path, then you will probably be pretty unpredictable already, wether or not someone else has the source code.

Re:Join Team Overbot - no pay, some risk, big priz

[Animats]

Going 120MPH, rather than 40MPH, for 20 miles of flat, hard plain (one likely section of the course shown in DARPA's pictures is such an area) gains 20 minutes. (10 minutes vs 30 minutes).

Going 30MPH, rather than 15MPH, over 20 miles of tough terrain, gains 60 minutes. (90 minutes vs 30 minutes).

That's why rough-terrain performance is more important than flatland performance.

The Polaris Ranger has a thick ABS plastic skid plate over the whole bottom of the vehicle.

But this isn't about the vehicle. It's a software and sensor problem. What's really needed to do this well is better laser rangefinders. They're coming.

John Nagle
Team Overbot