Slashdot Mirror
US Regulators Investigating Tesla Over Use of 'Autopilot' Mode Linked To Fatal Crash (cnbc.com)
An anonymous reader quotes a report from CNBC: The U.S. National Highway Traffic Safety Administration said on Thursday it is opening a preliminary investigation into 25,000 Tesla Motors Model S cars after a fatal crash involving a vehicle using the "Autopilot" mode. The agency said the crash came in a 2015 Model S operating with automated driving systems engaged, and "calls for an examination of the design and performance of any driving aids in use at the time of the crash." It is the first step before the agency could seek to order a recall if it believed the vehicles were unsafe. Tesla said Thursday the death was "the first known fatality in just over 130 million miles where Autopilot was activated," while a fatality happens once every 60 million miles worldwide. The electric automaker said it "informed NHTSA about the incident immediately after it occurred." The May crash occurred when a tractor trailer drove across a divided highway, where a Tesla in autopilot mode was driving. The Model S passed under the tractor trailer, and the bottom of the trailer hit the Tesla vehicle's windshield. Tesla quietly settled a lawsuit with a Model X owner who claims his car's doors would open and close unpredictably, smashing into his wife and other cars, and that the Model X's Auto-Pilot feature poses a danger in the rain.
Luckily for Tesla this sounds like it couldn't have been avoided in any way.
On the contrary, this seems like exactly the type of collision that auto-pilot systems should offer vastly improved protection from:
"Neither Autopilot nor the driver noticed the white side of the tractor trailer against a brightly lit sky, so the brake was not applied," Tesla wrote.
Seems like a pathetically primitive excuse for an "electronic eye" to me.
According to the article this was not something the driver could see and avoid, and while the autopilot could not see it either, based on the data from this crash it *could* see it the next time. Drivers learn from their own experience and fatal crashes terminate their learning experience, while autopilots learn from ALL autopilots on the road, and there are no "fatalities".
Of course it is always the fault of humans in the end, in this case the tractor trailer was not supposed to be there, so we'll only have perfect records when we get rid of all the drivers and have all the cars on autopilot.
Violence is the last refuge of the incompetent. Polar Scope Align for iOS
My opinion is that Tesla's self-driving system is not nearly as safe as they claim. One doesn't have to look very hard to find videos like this one where the driver has to react to prevent the auto-pilot from causing a crash. I question how long, realistically, a production Tesla can stay on the highway before a human needs to intercede to prevent an accident.
Given enough time, and enough lawsuits, I think that Tesla will shut off their self-driving feature. It needs to be a lot robust than it current is. I can't say with any expertise, but it seems like their competitors are taking their autonomous vehicle research far more seriously with plans to install a more sophisticated sensor package on their cars.
You might want to read up on the difference between compressive strength and shear strength.
What is the point of having a car with a self driving feature if you have to pay attention? Either the car drives for you or it doesn't. If it doesn't, the only way to ensure your fully involved in the drive is to drive.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Having the car on autopilot, but requiring the driver to "pay attention" and be ready to take over within seconds is the worst combination possible.
If the car did not have autopilot, the driver would have been more attentive, since he would have been driving the car.
If the car was completely on autopilot, then one hopes the computer would detect objects on the road, at least the bigger ones, like lorries and tanks.
What is the reason one would switch the car to autopilot? Most likely so that they can be less attentive to the road. If I need to be as attentive as driving an older car, then I will not use the autopilot. The reason is that the constant minor adjustments I usually have to make (the road is not straight after all) help me to keep my attention on the road.
If autopilot means that the driver has nothing to do most of the time, but has to react really fast when something bad happens is a problem, because boredom reduces the attentiveness.
Also, let's say the driver was paying attention to the road. The car is on autopilot, he sees that a lorry is getting closer to him. How is he supposed to know that this is the time the computer will fail to notice a huge lorry (it noticed much smaller cars and pedestrians with no problems before) and take over?
Why couldn't it see the tires, undercarriage, and side reflectors? And if the image was so washed out that it couldn't make out the outline, then because it couldn't see clearly, the car shouldn't have been moving so fast. It violated the Basic Speed Law just as surely as if it had been driving the speed limit in heavy fog, and that's a programming error.
It would also help to upgrade the camera to one with a wider dynamic range and/or more resolution so the image is less likely to get washed out again.
So there's a software fix and a hardware fix that will prevent this from happening again in the future. Unavoidable, my foot!
Any sufficiently unpopular but cohesive argument is indistinguishable from trolling.
The balance is that Tesla will learn from this accident. They will change the software on the existing vehicles to try to detect this situation better, and they will undoubtedly outfit the next generation of cars with improved sensors to avoid this specific accident.
In contrast, in a human-driven car, the only one who learned anything is dead, so the next person who gets in the same situation will likely react the same way and end up just as dead. At best, there might be a slight change to driver education because of it, but it isn't worth adding e.g. an extra lesson to the curriculum to avoid one specific accident.
Finally! A year of moderation! Ready for 2019?
Autopilot still has a far better safety record than human drivers.
There is insufficient data to take that claim seriously at this point. Autopilot features are promising but need a LOT more miles in real world conditions before it is safe to make generalizations like that. I know a lot of people have high hopes for autopilot (myself included) but let's not let our expectations get in the way of scientific evidence.
One thing I do feel comfortable stating is that more people are going to die before autopilot features become truly safe. There are lots of corner cases that we're going to have a hard time predicting and we'll only learn about after some accidents occur. This is the case with any new transportation technology. Airlines are very safe these days but early on they were significantly less so because there were problems we didn't know about yet. Airline windows are rounded because we learned the hard way about stress fractures from sharp cornered windows which wasn't obvious at the time. People had to die to learn that lesson. This won't be the last time someone dies making autopilot safe.