Slashdot Mirror
Stanford's Self Driving Car Tops 120mph On Racetrack
kkleiner writes with this snippet: "Just as Google's self-driving Prius goes for distance, recently passing 300,000 miles, Stanford's self-driving Audi TTS instead has the need for speed. The Audi, known as Shelley, sped around the Thunderhill Raceway track north of Sacramento topping 120 miles per hour on straightaways. The less than two and a half minutes it took to complete the 3-mile course is comparable to times achieved by professional drivers." Now if only Montana could take a cue from Nevada's rules for self-driving cars, and bring back "reasonable and prudent" speed regulation, driving out west could get a lot more exciting.
I would expect a computer-controlled car to do well in these kinds of situations. On a fixed course with no other cars, it comes down to calculating the optimal trajectories, and being able to accurately estimate things like when your tires are about to lose traction. Computers are probably better at that than humans are, given enough data. I mean, cars and tired are already designed with computer simulations of those kinds of conditions.
Google's self-driving cars being able to drive in regular traffic was more of a surprise to me: something I would've have expected for another decade.
10 PRINT CHR$(205.5+RND(1)); : GOTO 10
I grew up in Montana under the "reasonable and prudent" speed limit. Man, I miss those days. The problem was that too many tourists came in that didn't know the roads and got themselves killed, so the feds threatened to yank our highway money unless we changed the law. Unless you can do away with either the Federal government or idiot tourists, it's probably not coming back.
Yes, well it does seem rather ... intuitively obvious that it never actually exceeds certain death.
I have quite a bit of HPDE experience.
First off, quoting times around the track is silly unless it was in the same car. Which it wasn't. However, if you want to see what "fast" is, look at the SCCA records for various classes. Spoiler: lap times of 1:39 to 2:12. Read that again: the absolute slowest competitive race time is 2:12, and that was done by someone in a Mazda Miata in a stock racing class (ie, limited modifications.) The Stanford car has more than 100HP over the Miata, all wheel drive, big brakes, and a dual-clutch gearbox that shifts virtually instantly.
120MPH sounds impressive, until you realize that we're talking about a nearly 270HP car and a very open track. 120MPH isn't that hard to hit on many racetracks, even for a novice, and it's not a demonstration of skill; what's a demonstration of skill is how fast you exit each turn. Just by looking, I can tell you the fastest part of the track is between turn 8 and 9, most likely, for high-powered cars; slower, lighter cars may be faster between 9 and 10.
Second: "professional driver" could mean anything from someone who drives a taxi, to someone who races dirt-track, to someone who races Formula 1. Anyone can call themselves a "professional driver."
Third: the way that thing drives itself is absolutely atrocious and reminiscent of the worst kind of first-day HPDE students. The ones who think they know how to drive, don't, and are aggressive. Hammers it down the straights, not smooth with the controls at all, misses the apex (the inside center of the turn) by half a dozen feet, overloads the tires (hear them screaming? That's not a "I'm giving you the most grip" noise, that's a "I'm past my limit and am sliding all over the place" noise)...ugh.
From the way the car dives and rolls, as well as how the 'driver' is thrown around and the steering wheel is jerked - there is absolutely no finesse, and that is critical for driving fast.
Lastly: "For example, the math involved in getting a spinning wheel to grip the pavement is very similar to recovering from a slide on a patch of ice. "If we can figure out how to get Shelley out of trouble on a race track, we can get out of trouble on ice," Gerdes said."
Haha, no. Pavement, ice, dirt, and snow all have very different characteristics and "getting out of trouble" on them is different. Effin' Californians... Spend a winter in Vermont, then tell me about how to drive on ice.
Please help metamoderate.
For AI is the same, don't worry soon it will outrun, outpace, ... and make you completely useless. But it's ok.
On car topic: I can't wait for this to be a safety feature, Imagine, you drive full throttle down the road, but you miss-judged a corner, the AI, with light-speed reaction time understand that you can't make it at that speed and applies the little correction you need to make it through. It's the same, with traction-control, ABS, ... this will just be the next logical step.
Love many, trust a few, do harm to none.
I don't know the statistics for motor vehicles, but for bicycles, the common accidents are motorist-at-fault but avoidable by following best practices on the part of the cyclist (typically right-hook and left-hook, avoidable by things like proper lane positioning -- taking the lane rather than trying to ride in the gutter to avoid encouraging motorists to pass unsafely, using positioning to encourage drivers making right turns at an intersection to go behind rather than in front of you, etc) or cyclist-at-fault and thus avoidable (riding at night without lights, riding on the wrong side of the street, running intersections), and only a very tiny percentage are motorist-at-fault and unavoidable (ie. the "struck from behind while riding safely and properly" accident that everyone worries so much about... has a high chance of being lethal should it happen, but frequency is almost negligible).
That said -- I'm curious as to whether the parent's asserted statistics more correctly refer to the party at fault in lethal accidents as opposed to the parties harmed in accidents. My suspicion would be very much the former.