Tesla ‘on Autopilot’ slams into parked fire truck on freeway
mercurynews.com
mercurynews.com
The really scary ones are the ones that own it, know about the limitations, and intentionally bypass the safeguards.
https://insideevs.com/tesla-autopilot-survey-results-98-unde...
To be clear, these are from people who really like the feature and find it useful, but they are really aware that it has to be watched just like if they were 100% manipulating the controls.
If I made rat poison that looked like candy and named it “Yummy Candy Rat Poison” and some kid died coz they ate it, I’d deserve some blame regardless of parents keeping poison away from kids.
https://carbuzz.com/news/teens-take-tesla-model-s-for-joyrid...
I believe the problem lies in "Enhanced" and "Full self driving hardware". While this is an important point and is probably true, there should probably be bigger warnings that there is currently no full self driving software, and therefore, no full self driving.
Samething with an airliner, autopilot will not avoid another aircraft.
Wrong! There are autopilots which integrate TCAS into the autopilot (A380 among them).
Source: http://www.airbus.com/newsroom/press-releases/en/2009/08/eas...
Tesla Autopilot is a combination of both steering and throttle control.
This crash is unfortunate and should be investigated. But autopilot naming is something I don't think is the problem.
Pilots do so all the time, because airplanes aren't flying in a congested 2D space. Airspace is mostly empty, with very large distances between planes and features, allowing for plenty of early warning.
And an airplane's "autopilot" has nothing to do with Tesla's Autopilot. Their purpose and working mechanisms are completely different.
The airplane analogy is pure bullshit.
It operates the controls but relies on the human operator for supervision. Pilots talk about “flying in the magenta” and exactly the same thing applies to Tesla Autopilot.
Believing that autopilot means something different is the bullshit attitude that has to change.
Warren Van Der Burch gave a classic presentation, Children of Magenta: https://youtu.be/pN41LvuSz10
There are two people flying the plane, with rigorous training and controlled work hours. Air safety, as they say, is no accident.
Meanwhile a Tesla "autopilot" is designed to drive in traffic with other vehicles.
Tesla Autopilot can keep pace with surrounding traffic and stay between marked lines but does not recognise stationary obstacles.
They are very much the same in that they automate the mundane parts of the operation of the vehicle. They are also the same in that they do not know how to handle conditions outside their operating envelope.
The A380 begs to differ...
http://www.airbus.com/newsroom/press-releases/en/2009/08/eas...
Even pilots in IFR conditions are doing little more than maintaining a particular altitude and heading.
This means that a aircraft and vehicle autopilot have very different requirements. They are not, at all, doing the exact same things.
Then he failed, as they are completely unrelated.
> Pilots are never expected to get their eyes off the controls.
This is also false. They need to be alert, but they most certainly do not need to be as attentive as keeping their hands on the controls unlike with Tesla.
Unlike the sibling comment below alluded too, many commercial airplane autopilot systems are advanced enough to land the plane completely on auto as well as some even automatically respond to TCAS alerts to avoid nearby aircraft.
I do not believe Tesla’s are equipped with the necessary hardware to scientifically show beyond a reasonable doubt they assisted in preventing accidents.
Tesla has not been counting collisions caused by the driver having to brake hard because the autopilot was about to happily slam into another vehicle/object/person? Tesla's summation of those is that they were not a failing of Autopilot, because it was not active at the moment of impact. That's ... disingenuous ... at best.
And then:
"It would have avoided this accident [by braking, steering]", and studiously saying nothing about the future, "... but it would still have been in a collision 0.42 seconds later".
For all the collisions it would have avoided, there's another subset where it would only have "delayed" the collision. But that won't be mentioned. Because it doesn't fit the narrative.
These accidents are not people mistaking the purpose of autopilot, they are people slipping into false confidence that the car will handle the driving for them because it has in the past.
To be fair, an autopilot, in the traditional sense, just maintains a heading and an altitude.
Tesla wants autopilot
Emergency services want autonomy in being able to shut the freeway down.
How about PUT THE FUCKING AUTOPILOT SYSTEM IN THE EMERGENCY vehicles -- such that THEY can report to ANY auto driving car "SLOW DOWN we are doing emergency services [HERE]"
And then you get both -- The emergency services vehicles should be freaking beacons to any self/auto driving car.
To assume that this was short-sighted on either side is correct. Tesla should have been thinking about this, emergency services should have been thinking about this.
Why doesnt Tesla take their Semi model and make a Fire-Engine model????
Difficult to blame the emergency vehicles when the semi-autonomous car fails to detect something that stands out so much.
The automation in Tesla's vehicles is, and should be, designed around modern conditions, not a utopia. Otherwise disable all those features until that utopia arrives.
Or would we say collisions with pedestrians are okay until such pedestrians all have autonomous human beacons (made by Tesla, of course)?
This thread happens with every Tesla crash, by the way:
1. Tesla crashes
2. Driver says autopilot was engaged
3. Long debate on HN about exact meaning of word "autopilot"
4. 9 times out of 10, Tesla log says autopilot wasn't on before the crash, or that obstacle was visibly in the path for more than enough time for the driver to take action (usually half in the lane and half out)
The other one time out of 10 the driver says the Tesla suddenly accelerated and drove through most of a strip mall.
And what are stats of accidents on Auto-pilot vs without? If less, I'd say it's a success.
Have you ever driven a Tesla?
It's really difficult to stay focus and (mostly) do nothing at the same time. Monitoring automation is boring. Also switching from a passive mode to an active mode in few seconds in case of an emergency might be really hard.
I'm not sure the Tesla "autopilot" is designed in a way that really keeps the driver in the loop and active enough.
It sounds like it's not 100% confirmed that Autopilot was at fault. The first sentence of the article also reads "reportedly on 'Autopilot'"
Isn't that usually because the owner lied about using Autopilot?
The only thing that keeps a check on the impulses of the most aggressive human drivers is the unpredictability of other human drivers.
For example, if your computer control car knows that the car behind is also computer controlled, it could be more aggressive on the brakes to avoid a forward collision without causing a rear collision.
Tomorrow, the situation might be that a computer controlled car could see that the car behind is tailgating and the car in the next lane is computer controlled — then determine the safest action would be to swerve into the next lane rather than brake suddenly...
And I would dispute that emergency braking suddenly while being tailgated is less "crazy shit" than swerving into the next lane where your presence will, with an exceptional degree of confidence, be immediately detected and compensated for.
To clarify, the scenario in my mind is one that would be bold but not dangerous; something a good human driver might do with 360° situational awareness. In this instance it could be the difference between swerving to miss an obstacle or a multiple-car collision — even if your self driving car can stop in time, the tailgater is likely to slam into you, which in turn slams you into the obstacle.
Or to change the situation a bit, if the "tailgating" car happened to be a self driving car with an aggressive profile, that could swing the decision back towards braking. Point is, knowing that a car is being driven by a computer can aid decisions made by other cars.
By the way, one can speculate there are interesting ways to trigger beeps inside these Teslas coasting on the left lane.
It doesn't take a single car to reduce the speed of a lane. It takes exactly two. One to slow it down, and the other one to stop the others from letting the guy in the front know that he is going too slow.
Please drive safely.
*Big Red Truck
I wonder how Tesla's AEB stacks up against similar systems on other vehicles? There are a lot of vehicles on the road now with automatic braking, but we don't see every single frontal collision on such cars reported as some kind of systemic failure.
Why am I not already driving a Tesla?
The Tesla isn't magical, made of adamantium, or such.
Not to sound overly snarky, but as someone who has been a firefighter / paramedic for the best part of a decade, I've certainly seen my fair share of MVAs up close.
Airbags also kill babies if used incorrectly.
There are warning placards in several places in your car saying such.
Sometimes an airbag can cause more injury than the accident.
A seatbelt may trap you in a sinking car.
But overall, airbags and seatbelts have significantly reduced traffic injuries and fatalities.
Autopilot saves lives. Autopilot used incorrectly can kill people. But overall, Autopilot has significantly reduced the number of accidents involving Teslas.
So do we take airbags away because using them incorrectly might hurt a child? Are they a safety feature or marketing value?
There is no evidence of this.
http://bgr.com/2017/01/19/tesla-autopilot-crash-safety-stati...
Time will tell as they ramp production.
> "The probability of having an accident is 50 per cent lower if you have Autopilot on," said Musk, speaking at an energy conference in Oslo, Norway. "Even with our first version, it's almost twice as good as a person."
> The results are from Tesla's first generation of Autopilot, which owners have already used to drive over 47 million miles since it launched in October 2015.
I don't know what the answer is, but there's certainly evidence.
Sorry, I don't get it. How do you use something incorrectly when by definition you aren't really using it at all? It seems like an oxymoron.
Engineers are required to to make some kind of input to verify that they're still alert, or the train stops[1] if they haven't done anything in X seconds.
Obviously you wouldn't want the car to just stop in the middle of the road, but there could be something like that designed.
[1] https://www.quora.com/If-a-train-driver-falls-asleep-what-ha...
I'd imagine Tesla wasn't thinking about how the driver alertness feature could be countered unfortunately. Perhaps we'll see Tesla adapting the eye tracking tech that Cadillac Hypercruise is using.
I wonder how to truly tell if a driver is attentive? The Cadillac indicates it uses head tracking technology. I wonder if that is sufficient? Seems you probably need even eye-tracking: an attentive driver keeps looking around or something.
I suspect the primary problem is not the partial autonomous technology, but the marketing which is convincing people they can trust it. Nobody ever told you your cruise control means you don't have to pay attention, so very few[1] incidents happen due to misuse of it.
First and foremost, Tesla should never have been allowed to release "Autopilot" being called that. From the very name it provides a false impression of what it can do.
[1]I know you can find one. Humans are remarkably stupid.
It's true that people may be in particular need of a refresher, since such features could be perceived as more than advanced cruise control. It should be strongly emphasized to all drivers that we need to take full control whenever we see any exceptional conditions, from hazards, obstructions, pedestrians or wildlife in or entering the road, to poor visibility, and even when passing or being passed by small vehicles.
More than half a century in mass-market automobiles. https://en.wikipedia.org/wiki/Cruise_control#History
Tesla I think has done a fantastic job at striking a balance between getting something into the market the generate real-world data and mitigating the risks that come with it. If this car really did crash into the fire truck at around 65 MPH then it did a pretty good job at keeping everyone safe.
I think I would be hard pressed to find any technological or mechanical innovation that did not initially involve risk or have an accident associated with it. I've seen more examples where a Autopilot has done the right thing and avoided an accident than I have read stories of it making a mistake and causing an accident.
I believe your definition of keeping everyone safe differs from the common usage.
I don't see any evidence that Tesla has more, or less for that matter, of the necessary high-quality data that is needed for self-driving cars. The fact that real people are driving those cars continuously, potentially generating reams of data, is pretty immaterial, since the majority of that data is going to be useless (let's face it, 10 million hours of driving in the same conditions isn't better than 1 million hours of driving).
In my view, driving is already boring enough that it's insufficiently attention-grabbing for safety's sake. That's why, e.g., the "naked streets" model is removing lane marking. Decreasing (false) certainty forces more cognitive engagement, making things safer: https://www.theguardian.com/commentisfree/2016/feb/04/remova...
If anything, for the decades [1] while we're waiting for full automation, I'd like to see the equivalent of Threat Image Projection. It's the technology for luggage scanning where they put images of fake guns, bombs, etc in the images of real luggage. If the operator doesn't press the "threat" button they are alerted that they slipped up. It brings the frequency of important events up into the range where humans can pay attention. It'd be great to have the same thing for cars, so that AR images of deer, cyclists, children, and fire trucks appear; if you don't attempt to take evasive action, you lose points. [2]
[1] E.g., iRobot founder Rodney Brooks says not before mid 2030s: http://rodneybrooks.com/my-dated-predictions/
[2] I am, of course, kidding.
This type of accident in Level 2 systems is never going to be zero.
If the roads had sensors in the "bots dotts" and would track and steer all the volume of traffic - isnt that better than all the infantile-AI-cars doing it?
Pilot training covers techniques for avoiding automation dependency (aka letting autopilot kill you and all tour passengers). Perhaps we need the same for car drivers: if the car was as good a driver as you, it wouldn’t give you the option to drive.
I've heard of people not turning off their cruise control when going through an accident area. Boneheads. This is no different - just involving the thing after cruise control.
Will this turn out to be another "brightly lit sky" autopilot bug, like the 2016 semi truck incident?
https://www.tesla.com/blog/tragic-loss "Neither Autopilot nor the driver noticed the white side of the tractor trailer against a brightly lit sky, so the brake was not applied"
https://www.tesla.com/blog/upgrading-autopilot-seeing-world-...
> Because of the force of the impact, firefighters advised the Tesla driver that he should be taken for a medical evaluation, but he showed no significant injuries and refused treatment,
It is not allowed in Japan (where I live) because of the draconian vehicle inspection program, but I guess in some US states it is more liberal.
So was it on autopilot or not.
"Autopilot was not active at the time of collision, according to logs"... because it was deactivated by hard manual braking a moment before, because Autopilot was about to cause a hard collision...