Slashdot Mirror
Autonomous Race Cars
Octothorp writes: "Though not as complicated as the underwater
vehicles. There is an annual competition sponsored by National Semiconductors to build an autonomous race car. They move along pretty well too, at almost 9 ft/s. More technical information on how they are built is available on a Berkeley page, and there's a video of the winning run for 2002."
I recall an experiment done in California a few months back whereby cars automatically drive themselves on a stretch of a highway equipped with magnetic strips down the centre of the lane.
Since public transport seems to be out of the question for medium-distance transportation in the States (witness Amtrak's plight - and CoachUSA's financial gymnastics of late) automated private transport might fill the gap - provided price can be brought down.
It is really interesting to see what comes out of this. Certainly more applicable in the near future than football-playing robots - or this terrible series in the UK with robots battling each other, armed with saws, hammers and whatnot contraptions..
Curious,
Michel
Fedora Project Contribut
Linking to a video file from the frontpage of /.?
Wowzers! Methinks flaming shards of this server will be moving - autonomously - at 9 ft/sec very shortly...
[/bad joke]
The key to this particular competition seems to be the size and weight limits on the cars. This is a contest for little RC-style cars. The technology exists to go much faster, but not in this form factor.
It's my opinion that autopilot for your car is an inevitable development. It will become the only way to keep highways scaling. Adding more lanes to add capacity works to a point, but doesn't work forever. I don't think it's feasable to expect a human to navigate a twenty or thirty lane highway. And the density of traffic which an existing highway can carry is limited by the poor driving ability of humans.
I think eventually high-capacity highways will require the use of an autopilot. Doing so would allow the cars to be run with inter-car gaps which would be suicidal with a human behind the wheel. Most stop-n-go situations are due to bad planning on the part of drivers. They speed up too much when traffic clears ahead, zoom up on the cars in front, and then have to slow down to avoid an accident. This type of driving creates waves of congestion which travel backwards down the highway, and is due entirely to poor coordination among drivers. But there is no reason that under computer control rush hour can't cruise along at 60 miles an hour with a car length or less between cars. I bet you could easily triple the maximum capacity of a highway, not to mention getting everybody to their destination faster and with better fuel economy to boot. The R&D will be expensive, but like any electrics the hardware will be practically free once developed. Compared to the cost of expanding existing freeway's, it will make sense finacially too. I figure it's maybe 10 years out.
American's are not likely to give up their cars for any sort of public transportation, no matter how impractical cars become with rising fuel costs, increasing travel times due to congestion, increasing insurance rates, etc. Most Americans have convinced themselves that they enjoy sitting in stop and go traffic, as long as it's in a car and not a bus. But if we could figure out a way to let them keep their cars, reduce pollution, reduce accidents, let them safely talk on their cell phones, and not have to build mile-wide highways I suspect a lot of us might go for it.
The streets are jam-packed with thousands of vehicles travelling in irregular serpentine patterns.
None of these machines (except mine, of course) contain any type of human intelligence, but it's interesting to watch the AI at work. At night the roads look like Conway's Game of Life running on a computer with bad RAM.
Expert systems allow some vehicles to negotiate left turns from right lanes and to outbrake school busses when entering a rotary.
Fuzzy logic is essential for speed control, stop lights, parking and many other mission-critical tasks.
Genetic algorithms tend to select the maneuvers which are least expected by other vehicles.
Task scheduling is done according to driver convenience. For example, turn signals are always lower priority than dialing a call on the mobile phone.
Most communications between vehicles is a crude form of "digital" communication.
Unfortunately, most of the vehicles are Windows-based which results in a high rate of crashes. Mack trucks seem to be better than average.
With a good onboard computer, the car could build a model of the track as it went around, and calculate the optimal path & speeds to use on all subsequent laps. Using lasers or ultra sonic distance sensors would let the robot know when a turn was going to happen a lot sooner the the few inches of warning it gets in their setup. If you put an accelerometer in the car, you could even have it self-calibrate, discovering it's own acceleration curve, maximum lateral acceleration, and braking. It could then use those values to find the perfect path through the track on the second lap.
Yes, it sounds like a good idea, and in fact that was the original plan for the winning car. But it sounds a lot easier than it is. First of all, time constraint wise - these cars are built from scratch, and tuned to perform well in under fifteen weeks. (Yes, that includes all the sensor circuits, and power supply electronics) Secondly, there is a major problem with wheel slippage - if your wheels slip, you don't know where you are anymore.
A entry from 2001 went slow around the track the first time to memorize it, and then used that information on the second round in order to predict turns and change speed. It used the track crossing location to resync where it thought it was on the track. But if you look at the track layout, there are large section with no track crossing. i.e. wheel slippage - knowing where you actually are - is the main problem to be solved for memorization type approaches.
You can't see it in the video, but the winning 2002 car does detect and speed up (slightly) on straight aways.
Steve VanDeBogart
My son is building one of these. He gets the parts from a magazine subscription. Light following/repulsion, sonar detection, line following. And seems to go pretty fast to boot.
MrCreosote Meow!Thump!Meow!Thump!Meow!Thump! "You're right! There isn't enough room to swing a cat in here!"
The winner was a constant-speed car. At 6MPH. Cars aren't learning the track and then driving it at a good speed; they're just dumb line-followers.
Battlebots are much cooler.