Slashdot Mirror
Self-Driving Car Will Make Trip From San Francisco To New York City
An anonymous reader writes with news that Delphi Automotive is undertaking the longest test of a driverless car yet, from the Golden Gate Bridge to midtown Manhattan. "Lots of people decide, at one point or another, to drive across the US. College kids. Beat poets. Truckers. In American folklore, it doesn't get much more romantic than cruising down the highway, learning about life (or, you know, hauling shipping pallets). Now that trip is being taken on by a new kind of driver, one that won't appreciate natural beauty or the (temporary) joy that comes from a gas station chili dog: a robot. On March 22, an autonomous car will set out from the Golden Gate Bridge toward New York for a 3,500-mile drive that, if all goes according to plan, will push robo-cars much closer to reality. Audi's taken its self-driving car from Silicon Valley to Las Vegas, Google's racked up more than 700,000 autonomous miles, and Volvo's preparing to put regular people in its robot-controlled vehicles. But this will be one of the most ambitious tests yet for a technology that promises to change just about everything, and it's being done not by Google or Audi or Nissan, but by a company many people have never heard of: Delphi."
I was wondering how they planned to handle refueling (and maybe in the future, recharging). When they figure that one out, imagine what this kind of system will do to the trucking industry.
They don't even have to figure that one out. All they have to do is institute full service fueling to replace cardlock fuel stops. The trucks pull in, someone wanders over and fuels them, the trucks pull out again when they detect that the tank has been closed and the attendant has moved away. Extensive monitoring and high-resolution cameras will eliminate the need to have a human on board entirely.
Presumably, a car with a telepresence robot in it (just enough to communicate with attendants, not to drive the car) could conceivably already cross the country simply by stopping at full-service stations, if not for the legal impediments.
Besides commuting, OTR trucking really is the "killer app" for self-driving automobiles, and it's coming sooner than people think.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
More impressive would be for the car to drive from one end of New York to the other. During the day, avoiding highways, dealing with really chaotic traffic on narrow, poorly marked roads full of distractions and ambiguities.
Highways are simple. Traffic flows in one direction only, clearly marked and wide roads, no intersections, all roughly the same speed. No surprises. It's where by far the fewest accidents happen for human driven cars, even though it's boring and probably the part where human drivers pay least attention. Doing an hour of highways, ten hours of highways, 100 hours of highways - it's just more of the same. Now it's cross country, tomorrow it'll be cross country and back. And back again. As long as the fuel will last.
This is consistent with the overall American trend of replacing solid blue-collar jobs with entry-level service type jobs.
I fail to see how trucker is a "solid blue-collar job" while "gas station refueler" is somehow an "entry-level service type job". They're both pretty typical blue collar jobs.
I would say adding a full-service attendant at every truck stop gas station is probably the least complicated and easiest-to-implement part of an automated nation-wide self-driving truck shipping system. You're really focusing on the simplest part of the problem.
-- Let us endeavor so to live that when we pass even the undertaker shall be sorry. -- M. Twain
HaHaHa. Yes you can watch the semi slam on the brakes for a piece of cardboard that gets blown by the wind and creates a multi car pileup. I'm not sure how that helps.
Existing automotive driving assistance systems are already immune to this problem. They use radar to tell the difference between a plastic bag or a piece of cardboard, and something with mass.
Granted, a mylar balloon will probably produce the kind of results you're looking for, but there don't seem to be near as many of those on the highways as pieces of cardboard, so I can see why you didn't go for that example.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
I take it you have never worked on any computer vision papers or any teams that have tried vehicle automation? Because even autonomous freeway driving in the USA is far far far off - perhaps 30 years or more.
Cars with autonomous freeway driving will be out in just a couple of years, according to automotive manufacturers. Nearly all the major players are predicting fully autonomous cars will be a solved problem sometime between 2020 and 2025.
Keep in mind that cars can and will make use of a variety of sensors besides vision that are much easier and reliable to process. I'm betting the first generation of automated highway driving uses no vision systems at all... just radar, lidar, and sonar, plus GPS for nativation. More to the point, they can use ALL of them at once. Those are more than enough to handle highway driving. Cars today are already using some of these systems for their intelligent cruise control, auto-parking, and collision avoidance systems.
Make no mistake... computers are going to be FAR better drivers than humans. No getting drowsy or falling asleep. No distractions by the passengers in the seat next to you or the rugrats in the back. No rear-ending cars while looking at your cellphone or putting on makeup or one of the ten thousand stupid things humans do every day behind the wheel. Fully autonomous cars can't come fast enough, and they'll likely be coming a hell of a lot faster than you think.
Irony: Agile development has too much intertia to be abandoned now.
AI will not write books, do programming, etc. Strong AI is a myth.
Unless human brains have some magical powers (like a soul blessed by God), there is no logical reason that machines shouldn't eventually be smarter than humans. The only question is how far off it is. There has lately been some big progress on deep neural nets. There is also steady progress in symbolic AI and knowledge ontology. My guess is that we will get there in a couple decades, using a hybrid solution.
AI is just complex automation PROGRAMMED BY HUMANS.
That doesn't mean that it is limited by human intelligence. Good chess programs can easily beat the programmer that wrote them.