Slashdot Mirror
Google: Our Robot Cars Are Better Drivers Than You
An anonymous reader writes "At a robotics conference in Santa Clara, California, the head of Google's autonomous car project presented results of a study showing that the company's autonomous cars are already safer than human drivers — including trained professionals. 'We're spending less time in near-collision states,' he said. 'In addition to painting a rosy picture of his vehicles' autonomous capabilities, Urmson showed a new dashboard display that his group has developed to help people understand what an autonomous car is doing and when they might want to take over.' This follows another (non-Google) study earlier this week that found the adoption of autonomous cars could save thousands of lives and billions of dollars each year. Urmson also pointed out that determining liability for an accident is much easier using the data collected by the autonomous cars. At one point, a test car was read-ended, and the data showed it smoothly braking to a stop before being struck. 'We don't have to rely on eyewitnesses that can't be trusted as to what happened — we actually have the data. The guy around us wasn't paying enough attention. The data will set you free.'"
Have the Google robot take on the Stig round the top gear test track.
Autonomous cars will more than likely drive at exactly the speed limit. So on that stretch of highway you were used to doing 65mph in a 55 zone... well that slow car (hopefully in the right lane) will be the Google one.
I guess that's when the human takes over?
TODO: create/find/steal funny sig.
We'll soon reach a point where autonomous vehicles are orders of magnitude less likely than human-driven vehicles to have an accident. It won't matter, though; people would rather face a daily one-in-a-million chance of dying due to their own mistake than a daily one-in-a-billion chance of dying due to a machine failure.
Autonomous vehicles will still take over in the end. It's just that this particular rational motive to make it happen won't be contributing very much. So, it'll take longer than it should, and more people will die.
...in the future you are being looked at as being crazy if you tell other people that you are still driving yourself.
"Seriously, how can you live with that - risking the life of others. Robot-Cars are much safer."
Spelling errors were made for your amusement only...
if {collision}
then {arbitrary braking profile}
else {real data}
[Fuck Beta]
o0t!
If you had read TFA, you would have noticed that the robot car operates more safely than humans in the highway infrastructure that is in place today. We don't need to redesign today's infrastructure, if we switch over to autonomous cars.
Or imprison you, as the case may be.
Get it into production, allow for Moore's law, and these could be competitively priced in a very few years.
Thanks to our dear friends at the NSA, law enforcement will soon have the ability to override the destination selection of autonomous cars and have any driver/passenger they wish promptly delivered to a convenient jail or donut shop.
I love technology!
Scruting the inscrutable for over 50 years.
Solution: robot children and pedestrians. Anyway, wearable computing will make the garments aware of the surroundings. Trying to cross the road? Your pants know better!
That's all well and good until you're given a notice that they will be shutting down the automated driving service because they have Google Taxi now.
Demented But Determined.
You complain about "the expected cost".
Did you ever think about that EVERYTHING anyone earns anywhere is a "cost" for somebody else? Nature and economies are circular systems.
You WANT "costs" to be high - that means incomes are high. Of course, you don't want ANY costs to be high - battle tanks, mines, bridges to nowhere, poison gas, etc. are costs that are bad to have. Paying people to do nothing, by the way, is not on that category - these days A LOT of people would be much better paid to do nothing because what they DO get paid for is actually bad for the majority of people.
So "costs" are over all GOOD, but you have to look at the details, what they stand for. Too much abstraction is bad, comparing apples and oranges ("cost, money" makes everything seem completely equal) has gotten WAY too far.
Now go and take this out into New York City on 5th avenue at 5pm ET rush hour during the work week.
No, seriously, I want to see how well this car performs in a city where the posted 40mph speed limit oin the Staten Island Expressway is ignored by the vast majority of cops and motorists, the normal speed is about 70mph or so, and people will rear end you out of spite if you go too slow for them.
Then get me the data on how much less it costs to run this car.
you are so funny - you think drivers (or you for that matter) are some magical being that can predict actions?
Yes there is and it's already on high-end non-robotic cars. There are IR sensors that can see deer/bicycles/pedestrians even if your headlights don't pick it up yet at night or if they are (partially) hidden behind a car. Also, a robotic car can respond in a matter of microseconds, the human brain easily takes up a second to respond in these situations.
Custom electronics and digital signage for your business: www.evcircuits.com
1. Will you still be drving drunk if you have your autonomous car drive you home after a night of drinking? 2. What if you are driving link and ass and rear-end someone, will they be able to use that data against you? What if both people are at fault? 3. Who's going to absorb the liability for these cars when something unexpected breaks? The large automotives are going to drag their feet for years on self-driving cars. Their will need to be a lot of testing in real life before they mass produce any cars.
An autonomous car will be even better than you in that situation, *if* it's programmed for it.
One of the beauties of autonomous cars is that they can be programmed from simulations. Not only that, if these auto-cars have black boxes, then data from accidents can be analyzed and used to update other cars.
When our name is on the back of your car, we're behind you all the way!
So when you're driving today you're in a state of being aware of the situation and are engaged with the surroundings.
If you're letting the car drive, I highly doubt you're paying that much attention. Why wouldn't I let the car drive and I read, do email, surf the web or turn around and talk to the passengers in the rear seats.
In the event where you need to take an emergency action, it's much easy in the first case to go to heightened state than in the second one. Atleast in the first one you aren't completely surprised by the events you're facing before you.
Think of the case of a gravel truck that has a loose load. If I know there's a truck in front of me, I'm not 100% surprised if some gravel comes out, whereby if i'm reading/emailing and I'm forced to take over to avoid gravel, it's more of a surprise and I'm forced to figure quite a bit more out about the situation before I can act. One could also panic because of the amount of elevated emotion or adrenaline dump that would be taking place since you'd go from "reading iPad" to "dodging gravel".
There is no tech yet that can anticipate a child about to kick a ball out onto a road, or to see that a pedestrian is about to walk out in front of you without looking first.
Do you really think you 300ms senses are better at detecting 'random_object_in_car_path()' and doing a 'controlled_break( distance( random_object_in_car_path() ) / car_speed() );' than a laser detection system operating at sub-millisecond speeds?
Conservatism: (n.) love of the existing evils. Liberalism: (n.) desire to substitute new evils for the existing ones.
I'm pretty sure that the cars are constantly watching in front, back and both sides (too lazy to double check). They have been tested accident free (or at least fault free) in San Francisco, and successfully navigated Lombard Street. There is also driverless technology to allow them to see traffic (possibly just driverless traffic) around blind corners, obscured by buildings.
EVERYBODY says they are a good driver and better then others. So why would Google do otherwise?
I also like the bit at the end: The data will set you free.
Also : yes, you need eye witnesses. Or at least external experts.
I would not trust a company saying they are innocent in an accident and back it up by THEIR data. "We promise there was no software error in ANY of the cars. All people need to do is sit on the left cheek and hold the doorknob with the right hand. People just are using it wrong. They also signed a waiver when they opened the door. Look it up. It is in the Company-Is-Always-Right law that was passed last week."
Don't fight for your country, if your country does not fight for you.
Last week I had an American friend over and we were talking about driverless cars, and she said she thought they might work in the USA, but having seen what UK roads are like, she was very skeptical they'd work there, so maybe Google should try it!
For example, many roads in tows date back to roman times, and are too narrow for two-lane traffic. You need to look far ahead and work out when exactly you need to duck into a gap behind a parked car to let oncoming traffic through, and when to go for it when you have right of way so as not to block traffic in either direction. And if a block does occur, will it mount the pavement (sidewalk) to free things up, or know when it's time to back up and give in?
The UK has very few towns laid out in a grid, and most roads are twisty, and narrow, other than motorways. Can a driverless car cope with such terrain? If Google really want to prove their technology is better than a human, let them bring their cars over to the UK. If they work here, I'll be impressed.
Well up to a point. The comparison is with 'normal' drivers, American drivers at that. As far as I know driver training in the US is fairly basic compared to many countries, and TFA doesn't actually mention 'professionally trained' drivers at all unlike the /. summary.
I would contest that a 'real' advanced driver, say a UK police Class 1 driving certificate holder, will out-perform a robot any day, not because they are quicker to react but because the skills and experience they have gathered over years of dedicated training and practice would give them the ability to predict problems and prevent incidents, rather than relying on their robotically quick response times to cope with a developing problem the advanced human driver would have already avoided.
Smivs on the intertubes!
No, but the cameras could be equipped to look into the IR spectrum and see a child (or dog or other such heat profile) and attempt to determine whether its moving into the street or not.
Can never predict 100% of everything of course and even if you predict it you might be too late to prevent it but the computer's got far far more sensory input (potentially) available and is significantly faster at making mathematical calculations such as "trajectory of that IR blob that's behind the vehicle." than any human.
At the end of the day, there's not really any question about these vehicles having the potential to be safer than human drivers -- its basically all about the software implementation. If they can do enough stuff and handle enough special cases in the software, then as a human you will simply be outclassed in every situation except one -- having a "bad feeling" and stopping for only that reason. I would hazard to say that the number of times that's used as a successful accident avoidance mechanism pales a lot in comparison to the more common scenario of just not noticing something in time.
This already exists. On the way out of work, call a restaurant on the way home. Say "I want to place an order for carry out".
Many places walk it out to your car.
Expensive, but adds almost no time to the commute, and no cooking.
Having worked on self-driving cars (2005 Grand Challenge), a few points:
The comment about minimizing "near-collision states" is significant. A near-collision state is one where a reasonable variance of the behavior of another vehicle could cause a collision. It's about predicting other-vehicle behavior. That's an important area to study. Aviation people put a lot of effort into minimizing near-misses, and it pays off.
Incidentally, Tesla's announcement that they're starting work on an "autopilot" is them playing catch-up. Audi, BMW, Cadillac, and Ford are already demoing automatic driving systems. It looks like Cadillac will be the first to ship hands-off highway driving, in 2015. All these early systems are highway driving only, although Cadillac includes stop-and-go driving in traffic jams. That's likely to be a very popular feature.
On the sensor side, more progress is needed, and it's coming. That rotating LIDAR contraption on top of Google's self-driving cars is from Velodyne. It's 64 LIDAR units on a spinning turntable. That's a research device, not a production one. There are better ways to do LIDAR, but the cost needs to come down. The approaches used in the Kinect and the XBox One will not work outdoors in bright sunlight. Outdoor LIDAR systems work fine, but they're pulsed, not continuous. For a nanosecond, at one frequency (color) they far outshine the sun. But the total energy per pulse is low, so they're eye-safe. Currently, such devices are very expensive, but that's not for any good reason. It's because some exotic ICs have to be made in tiny quantities.
Radars are getting better, too. A decade ago, in the Grand Challenge, we had to use Eaton VORAD radars, which operate at 24GHz. These could reliably range cars, trucks, and larger bicycles, but not people at long range, or signposts. (Such radars return range, azimuth, and range rate; this isn't a speed gun. I used to have one of these looking out my window at at an intersection, with a display plotting the traffic.) Today's automotive radars are running at 77GHz, with plans to move to 79GHz. There's an effort to standardize on 79GHz internationally. Tripling the frequency, plus applying more compute power to the processing, means that most objects a car might hit are detectable. These radars are getting cheap and small, so a car will have enough of them to provide full-circle data. Long range is needed mostly in front; on the side and in back, much lower power can be used.
A key issue is a high viewpoint. This isn't just about obstacle detection. You also need to profile the road. This was a big deal for the off-road DARPA Grand Challenge, but even on paved roads you need to be able to detect junk on the pavement and potholes. Google has their sensor on top of the roof. This will probably be unacceptable in a production car. I'd go for flash LIDARs at the top corners of the front windows. One possibility is a narrow strip just above the windshield, to contain all the sensors. This is one way to combine vehicle aesthetics and field of view.
Cameras are useful, but computer vision is still kind of dumb. Distance from stereo only works at short ranges, and range rate info from cameras is poor. Digital cameras are so cheap now, so it's tempting to think they can do the whole job. Not yet. Computer vision isn't good enough. Tesla is probably putting too much hope into camera processing. You need cameras to recognize signs, traffic lights, and such. Also, you need multiple sensors because not all objects are visible on all sensors. Radars can't see insulators. Cameras can't see objects with little contrast against the background. LIDARs can't see some materials, such as the charcoal fabric used on many office chairs. Sensor fusion is essential.
Enough for now. This looks quite do-able.