Slashdot Mirror
The Documents From Google's First DMV Test In Nevada
An anonymous reader writes "IEEE Spectrum contributor Mark Harris obtained a copy of the DMV test Google's autonomous car passed in Nevada in 2012 and associated documents. What has not been revealed until now, is that Google chose the test route; that it set limits on the road and weather conditions that the vehicle could encounter; and that its engineers had to take control of the car twice during the drive.
All of the recent articles about autonomous cars seem to be trying to make people think they're terrible will never work and are a disaster waiting to happen.
All of these tests and such aren't being done so they can release an autonomous car tomorrow, its an ongoing process and will take time. I don't blame google for not wanting to publish all the details about it, its a research project and the media seems to have an agenda to make autonomous cars into the boogeyman
Yeah, and it went into manual controlled mode when it recognized obstacles it couldn't handle:
A. A railroad crossing without signals
B. A roundabout
C. Construction work
D. "Some specific turns"
Obviously not ready for the real world yet.
From TFA: 'In communication with the Nevada DMV before the test, Google said its policy was to prohibit autonomous operation at railroad crossings that lack signals and for human drivers to take over. It also noted: “[Roundabouts are] particularly challenging, where many drivers don’t know the proper rules in the first place.” In an e-mail to colleagues at the DMV, Breslow wrote, “We can’t fail an applicant for not being able to navigate a traffic circle if they say that there [sic] vehicle can’t yet do it.” '
So the two times that the Google engineer took over were for the two things that Google said they felt it was unsafe for the car to handle - a railroad track without a signal, and a roundabout.
Occasionally living proof of the Ballmer peak.
Ofcourse it is not 100% ready for the real world. It does not mean it should not be deployed though.
Not every technology and device that made it to market first worked with 100% accuracy the first time.
It just means that drivers should still be paying attention to the road at all times instead of being distracted behind the wheel. The faster it comes to market, the more quickly the improvements can be made to advance the technology.
- SK
Why does version 2.0 of your application still not have all its features?
Because code takes time. And you can't just manpower your way through it.
Part of driving is dealing with a certain percentage of drivers who either don't know or simply fail to follow the 'rules'. But, I can see why roundabouts, particularly multi-lane ones, would be difficult to program the algorithms for. Part of effective roundabout driving is a bit more anticipatory than many other driving tasks.
The only slight problem with that is that in order to react at all in time, you must be paying the same amount of attention as you would if there was no autonomous drive system at all. This is otherwise known as the human being in the loop. Removing the human from the loop in aircraft automation has been a source of unending problems, and only recently one could say that it's a reasonably well understood problem - if not quite solved just yet. Don't forget we're talking about trained professional pilots here.
So, when faced with a self-driving car, the relatively untrained non-professional driver will always be so far out of the loop, that there's no way for him to overtake control safely in real time.
Of course, the solution for that is simple: the car's control transfer must, by default, happen in a fail-safe state - with the car stopped, with emergency blinkers on, etc. Only if the control transfer is explicitly acknowledged in a preset time, would the fail-safe be bypassed.
A successful API design takes a mixture of software design and pedagogy.
"engineers had to take control of the car twice during the drive." You mean like many of our parents did reflexively when we were learning how to drive? Not to mention the dent in the passenger side floor where the extra brake pedal wasn't.
"Win treats sysadmins better than users. Mac treats users better than sysadmins. Linux treats everyone like sysadmins."
I think the part about other drivers was the dmv claiming that failure to navigate a roundabout is not grounds for failing a human applicant. This report does more to illuminate how low the bar is set for any driver, machine or human.
Commercial aviation is now safer than it ever was in the past.
Fully autonomous driving is doable IF it is only along routes that have been verified and to some extent instrumented. I predict we'll see a few Approved Routes initially, such as stretches of Interstate. Fairly soon, the approved routes will account for the majority of vehicle miles driven. Then there will be a long tail of routes and conditions that won't be automated anytime soon. Basically, just like cellphone coverage.
Why does version 2.0 of your application still not have all its features?
Because code takes time. And you can't just manpower your way through it.
A roundabout is simply nothing more than a right hand turn at a yield sign, followed by an exit ramp. If your application can't handle that, then why the hell is it on the road in the first place? And what connection does this have to what *other* drivers do?
I have seen perfectly sober drivers go the wrong way on a roundabout. Self driving cars in a beta if not alpha program having to contend with asinine drivers like that would understandably need human intervention remember this is not a finished product, not even a release candidate, it is a limited alpha release it will be improved by the time it gets to market.
---Saying gnome 3 is better than windows 8 not so much a compliment as it is damning with light praise.
And let's just remember that planes don't actually fly themselves:
http://www.askthepilot.com/que...
-Chris
Along the same lines, it doesn't seem difficult to take control of the system while it's actively driving. It's not hard to disengage cruise controls or stop a car using Park Assist or Lane Assist from turning into something not seen by the sensor system. Why is it hard for me to grab the wheel from the "hands" of the auto-pilot in the Google car?
Here's the best example I can think of -- let's say you are the understudy for a radio actor with narcolepsy. You both have the script, you the understudy are following along word for word as the actor is performing. Suddenly the actor falls asleep and the words stop. How many seconds pass before you pick up where he left off? You are as aware and able as you can possibly be without actually anticipating something you can't anticipate, and I believe it would still take me a few seconds to switch myself from simply paying attention to audibly reading words.
Second scenario is the same except that you are both in sound booths reading the words and the actor is the only one with a hot mic. I believe it would be faster for me to be reading aloud with the actor and trigger my mic to go live at the necessary time -- however in doing so I am saving no effort over doing all the reading myself in the first place, so the application to automated vehicles is somewhat limited.
Ofcourse it is not 100% ready for the real world. It does not mean it should not be deployed though.
We need the power they said, it will be fine they said, don't worry they said.
The citizens of Chernobyl
Interesting analogy, since the Chernobyl accident was not caused by the power plant's automated systems, but by human beings that overrode the safety systems designed to prevent just such an accident. Interestingly, the Three Mile Island accident occurred for essentially the same reasons: humans prevented the automatic systems from functioning correctly to prevent an accident.
You're making the assumption that any situation the car cannot handle is both an immediate danger, and a situation that can be handled perfectly by a human.
When I try to think of situations where an automated car would fail, most tend to be ones where a response of "come to a full stop, don't do anything until the human orients himself and takes charge" is a perfectly valid one. Traffic lights not working? Let the human figure it out. Bridge out ahead? Let the human figure it out. Conditions so bad you can't see the road markings? Let the human try to do better, and if he wants to sit on his ass until it clears, that's probably a good idea anyways.
Sure, there are situations where an AI might not be able to avoid an accident an alert human would. Let's say a trailer detaches from a truck in front of you, but not in your lane. As it skids, it suddenly tumbles into your lane. An automated car might have ignored it until it was too late, while an alert human would have slammed on the brakes as soon as they saw it.
But how many humans would have been that alert? Even if they weren't on a phone, or sipping their coffee, or fiddling with the radio, most drivers end up in a sort of trance, doing things automatically. I've seen people crash just because they weren't paying attention - not distracted by anything, just driving without conscious thought. Automated cars won't have that problem - they don't *get* bored. Even if they can't dodge a freak accident, they'll be avoiding plenty of routine accidents. Net gain for people who don't like car wrecks.
The only slight problem with that is that in order to react at all in time, you must be paying the same amount of attention as you would if there was no autonomous drive system at all. This is otherwise known as the human being in the loop. Removing the human from the loop in aircraft automation has been a source of unending problems, and only recently one could say that it's a reasonably well understood problem - if not quite solved just yet. Don't forget we're talking about trained professional pilots here.
So, when faced with a self-driving car, the relatively untrained non-professional driver will always be so far out of the loop, that there's no way for him to overtake control safely in real time.
If you read the article, in the instances where the automation didn't know what to do, it pulled over and stopped:
Construction work, however, proved trickier. When faced with a partially blocked-off road, the car switched between autonomous and manual modes and then braked to a halt, requiring Urmson, the safety driver, to take control.
The driver doesn't need to react in time - the car does that. The driver merely needs to make the next decision to start moving again and guide the car to where it needs to go.