Tesla footage of braking before crash ahead [video]
youtube.com
youtube.com
Man: Wacht even wacht even eerst kijken wat er achter ons gebeurt (Wait, wait, first take a look what is happening behind us, don't get out yet.)
Man: Blijf zitten blijf zitten. (Stay seated, stay seated.)
Woman: Niets doen niets doen niet eruit gaan alsjeblieft. (Don't do anything, don't do anything, don't get out of the car yet please.)
Man: Nu kunnen we kijken wat er aan de hand is. (Now we can check out what happened.)
Woman: Maar zet hem eerst even aan de kant. (But first put it [the car] on the side of the road.)
We now see the man running to the car in a blue shirt on the left side of the screen.
Woman to the kids: Jongens jullie moeten even in de auto blijven zitten. (Guys, you have to stay in the car for a little while.)
Kids: Ja. We willen er ook niet uit. Moet jij ook blijven zitten? (Ok. We don't wanna go out. Do you [mom] have to go out as well?)
Woman: Ja ik moet er uit, ik ben een arts. (Yes I have to go, I'm a doctor).
Edit: just the first part of what the guy says- not the part about being a doctor from the woman.
It might help if you are willing and have the time (and it's enabled).
I'm not sure how to go about it though...
* The car behind didn't show turn signal so probably driver wasn't paying attention.
* A small car escaping with not too much damage while SUV rolls over many times
* From physics point of view, the car behind produced almost perfect torque that started horizontal spin on SUV
* Collisions with barrier at high speed can produce lots of rolls. Many time we see cars upside down in accident and wonder how did that happened. This is how.
* The middle barrier did its job wonderfully even against massive inertia of SUV. Cars on other side remained unscathed!
Living in India, I have seen a fair share of accidents where the car is upside down. I always used to wonder how that happens, now I know.
Is it purely because it hit it at the corner, so all the force went into the "spinning axis"? (did I say I feel dumb?)
Both cars braking, center of mass shifting toward their front and lowering. Meaning the small car momentum was peak at the collision point, while the SUV rear end was already lifted thus lighter.
Direction of the collision helped transfer the energy into torque. The barrier provided the final help to start the car tumbles.
A) main that SUV happened (just before starting sommersaults) to position itself orthogonal to its previous way of moving. In this position all wheels of SUV created massive 'resistance' to continue move forward
B) it happened partialy because of the middle barrier. WOuld it be missing, SUV would probably move (on the arc) without sommersaults.
Change of orientation was caused by small car, but look that up to moment A) SUV more less was moving fine.
Physics would have you start with a spherical, uniformly dense car, but cars aren't that. Crumple zones engineered for passenger safety requirement mean that intuition falls short unless you really dig into it and add stuff like center of gravity to the model.
Also think about why you believe that - looks can be deceiving, and marketing, especially for rare expensive purchases like cars, very much plays on emotional appeal rather than actual science. ("Ultimate driving machine"? Pft, really?). The original SUVs were designed off truck platforms and did quite poorly in safety tests, and their higher center of gravity doesn't help at all. (Though it looks like a minivan in the video.) Just like two objects falling will fall at different speed in the real world due to air resistance, bigger != better with cars. You can't see crumple zones just by looking at how pretty the car is; vehicle safety is one area where government regulation has made cars safer by requiring all new cars sold in the US to meet a very high safety bar, and that's improved over the years, evident in this video of an intentional crash between a 1962 Cadillac vs a 2002 model. https://youtu.be/O-WYKYrq5FI
r X F
meaning that small force applied over a long length can produce huge torque! You can move a huge rock if you use a long enough metal stick for leverage. This is why Archimedes said,
Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.
Large angular acceleration due to large torque is then directly result of Euler's equations.
Watch the video again; the little red car turns on their blinker and does a very slow shift to the right before clipping that stopped SUV. You're right that he wasn't paying attention, but his reaction time is incredibly slow - it looks like that car had just enough time to slam on their brakes or swerve to the right, yet they did neither.
aren't you supposed to have a wide enough space between you and the car in front of you? to compensate for any slowness in reaction.
I think the situation is not entirely unclear, it seemed that there was a problem ahead of the SUV, that made a bunch of cars slow down altogether. I still don't understand the smaller red car driver decisions.. maybe he didn't know what to do and hesitated and tried to swerve as a last resort. Or maybe he was just not watching. The SUV break lights were on quite a long time..
Most cars just don't flip, all the traction their tires have is simply not enough to do it on flat ground. Yeah anything could flip over once you go off the road but the SUV in this looks like a rolling death machine.
This is one of the reasons I ride bicycles and motorcycles. I take on more personal risk, but the overall effect is a reduction in accident, injury, and mortality rates for road use.
1. Hatchback hits on tangent producing torque and spinning SUV by 90-degree.
2. SUV collides with barrier which produces force vector that is normal to collision surfaces and slightly off forward direction.
3. The driver now seems to have pressed on breaks really hard which produces another force vector. But this vector is now quite off original forward direction because car had been rotating.
So the moral of the story is that if SUV is rotating and you press on breaks really hard, you will get roll. This is purely because SUV has higher clearance and so even small force can produce enough torque that will cause roll. It looks to me if driver hadn't pressed on breaks really hard, the rolls could possibly been avoided or reduced.
If you're referring to the pink hatchback that ultimately rear-ended the SUV - the pink hatchback seems to be making a lane change to the right after passing. The driver seems to have signaled and looked to the right to clear his/her blindspot at the same moment the SUV ahead slams on the brakes. It really seems like bad luck, that traffic would suddenly stop at the exact moment the driver of the pink hatchback looks to the right to change lanes.
Back in the day when I got my driver's license we were told to have 3 seconds of distance. One second to react one second to brake and one second for safety, if I remember correctly.
-Pet peeve of mine: "Bad luck" is incredibly rare in traffic. This accident occurred because the hatchback was too close to the car in front of it; not because of bad luck.
You're just missing the signal. Look at the upper-right of the hatchback at 4 seconds.
We did a lot of safety research before buying our current round of cars, and you need to not generalize about cars based on classifications like "SUV". Look at actual data from actual cars on actual roads, e.g.
http://www.iihs.org/iihs/topics/driver-death-rates
Note that you can find vehicles like the Honda Pilot that are enormously popular and have in effect ZERO fatalities. What's horribly unsafe to everyone is passenger cars built on truck frames.
Have any examples? The vast majority of new SUVs (which the one in this incident appears to be) are unibody rather than body-on-frame ("truck frame" style).
Apparently there are 18 models of Body-On-Frame SUVs that are being sold in the US in 2016:
Cadillac Escalade
Cadillac Escalade ESV
Chevrolet Suburban
Chevrolet Tahoe
Ford Expedition
GMC Yukon
GMC Yukon XL
Infiniti QX80
Jeep Wrangler
Land Rover LR4 *
Lexus GX460
Lexus LX570
Lincoln Navigator
Mercedes-Benz G-Class
Nissan Armada
Toyota 4Runner
Toyota Land Cruiser
Toyota Sequoia
You should never leave your children and family on a car on the left lane while you leave, never.
I have a neighbor that is paraplegic because of that. His entire family went to the hospital after a vehicle crashed from behind after a minor accident.
Your first priority is putting your car and family on a safe place. Proper triage priority number one is avoid another accident, by removing obstacles and signaling other cars.
they clearly said in dutch to wait and see what the traffic behind is doing.
Here, when accidents like these occur, our automated road information system directly displays how fast you may drive there (it has sensors to monitor traffic speed). And, within minutes, a traffic monitoring group will close down the affected roads fully and signal to which lanes people have to switch to by using arrows pointing towards the lane next to it.
The system looks like the following, and is placed on almost every highway here: https://upload.wikimedia.org/wikipedia/commons/8/88/Amsterda...
Not just for car accidents, but this is true for emergencies in general. Always check your surroundings!
For example, if you find a guy laying on the sidewalk, he could have stepped on a down power line, slipped on ice, or recently gotten attacked by some punk who might still be nearby. If you don't know why he is unconscious, proceed with caution.
Why would Autopilot have a warning for frontal collisions but not for rear collisions? And why would Autopilot be speed-limited and all that if it was able to use launch mode to avoid a rear collision?
Does anyone know if these braking systems account for people driving behind you? And if a collision is inevitable, what do they "choose"?
I would think in theory they could use the camera alone to detect an approaching vehicle, and take some remedial action.
And would this reduce the impact from behind if it were to happen?
That's the decision I had to make one day. I saw a car coming to hit me from the rear. I took my foot off the brake and gripped the steering wheel, leaned back and relaxed. It worked out well, but I've always wondered if the brakes make any difference one way or the other.
Do that a few times and even the most persistent tailgater gets it and either changes lanes or stops riding your ass.
Great habit, but after seeing over 10,000 charts for chiropractic cases (was working on IT software to categorized it) I learnt another trick that many doctors tell their patients (unfortunately after the fact):
Once in a emergency breaking situation, before you look into the mirror, secure your neck/back of your head as hard as you can into the headrest cushion. This trick will "unite" your head with the seat and in rear-end, it will push you and the car both at the same time, instead of - from what I seen in 80% of Chiro cases - your seat head cushion hitting your head in a moment when its pulled hard to the back due to G-forces related to being in a rear-end.
I have to say seeing some photos I quickly got into a habit of first doing that, next checking mirror - if I assume someone might not make it and I still have room upfront, I go forward as fast and as close as I securely can.
Please don't do this in the snow or in wet conditions. Sharp actions are what will make other cars change direction or magnify the effects of drivers that you have caused to panic. Putting on your hazards and slowing at a deliberate pace is safer for everyone behind you.
On freeways, each lane tends to develop its own speed, and at 4+ lanes the cognitive load starts to overwhelm me, especially when the relative speed difference between neighboring lanes is very high. At this point anyone in a slow lane may decide to try to pull out into the fast lane, or someone from the faster lanes might try to squeeze in to the queue of waiting cars; meanwhile if you try to unilaterally lessen the speed difference, the people behind and around you will get antsy and aggressive, often defeating the point.
These situations are some of the most dangerous that most driver experience: high speeds, large discrepancy between speeds, and limited escape paths.
This is why.
Driver errors with semis on the highway kill thousands every year. This kind of technology will save lives - not just of the Tesla drivers (or other autopiloted drivers), but of the vehicles they don't hit.
I totally agree with you.
It's similar to icy roads. If you go the normal speed limit and you slide into another car, it's your fault. The fact that the speed limit was higher than your speed doesn't release the driver from responsibility.
It's the responsibility of the traffic engineers to make sure the yellow light is long enough to accommodate reasonable braking time and distance for all of the traffic that can be on the road, travelling at the posted speed limit. They can lower the speed limit, increase the yellow light duration, or put up a sign limiting the max vehicle weight.
It's not reasonable to expect every driver to know detailed information like yellow light times for every intersection and section of road.
It's a big enough problem that there's a group trying to do something about it.
http://shortyellowlights.com/standards/
EDIT: according to the link, the yellow light for a 55 mph road should last 6 seconds.
Chicago was caught making their yellow lights purposely short on some intersections to make more money.
Given that it's all math at that point, I wouldn't doubt if regulations surrounding minimum yellow-light timing are updated to more accurately represent the stop time of heavy loads. Otherwise it should be easy to prove in court that stopping in time would be impossible given the speed limit & weight restrictions (or lack thereof) in place at a given intersection.
But in all seriousness, although I understand that people like to out-cynic each other, over the road trucks seem to be pretty well-maintained. Whether that's the result of safety regulations and enforcement or simple economics (a broken-down truck is not a moneymaking truck) is left as an exercise to the reader, but either way you do not see a lot of barely-making-it or broken-down commercial OTR trucks on the road. That level of poor maintenance is generally the province of private vehicles.
That said, I agree there's plenty of incentives for the trucking companies to go the autopilot route.
Fatigue and sleep deprivation are chronic problems in the trucking industry. Clearly, many drivers are unable or unwilling to maintain themselves.
https://www.nytimes.com/2014/06/17/business/truckers-resist-...
http://www.huffingtonpost.com/entry/congress-made-trucking-d...
They're almost incentivized for this change.
Out of 33883 US traffic deaths in 2009, 3380 involved trucks, and truck driver error was listed as a cause 26% of the time.
Self driving long-haul trucks probably won't save "thousands" of lives, but certainly hundreds.
http://www.trucking.org/ATA%20Docs/News%20and%20Information/...
For example, in the original video, if the Tesla autopilot hadn't braked, and the car had been involved in the crash, it's not clear to me that error on the part of the Tesla driver would have been listed as a cause.
Automating cars will take much longer than trucks.
Car #3 (Tesla), if it didn't stop, would be at fault for not maintaining stopping distance, hitting Car #2.
If Driver #3 can't see Car #1 past Car #2, and Car #2 is traveling at highway speed, I think it's a stretch to say that Driver #3 could be faulted.
Good. Some things to keep in mind: Semis can clear obstacles you probably can't - and at least around here are trained to not break because of some road hazards. Semis are not manuverable, and if they lose a tire the last thing they want to be doing is applying a now-uneven breaking force.
(sorry, link in German) http://www.spiegel.de/politik/deutschland/anschlag-von-berli...
This seems to be a fresh news story but I hope it will be more widely disseminated. Among so much aimless hand-wringing over terrorism, security theater "designed to stop it", and dragnet surveillance and intelligence gathering, here is a clear example of a new technology reducing the impact of a terrorist act and saving lives.
It kinda goes against the click-generating narrative of autonomous driving technology being half-baked gimmicks rushed to market, fatal accidents waiting to happen, and Tesla and its celebrity corporation ilk flying too close to the sun on homemade wings.
A too common practice in Europe is to drive by ear. The truck driver will not watch the road unless the wheels are running over the side lines which makes a noise.
It shows, how the radar detects, that the lead car breaks first - watch how it gets highlighted in the dashboard. This ability helped the Tesla to handle this accident event.
My read on what happened here is that the pink car was making a passing maneuver, and just as the pink car driver checked their right-side mirror/blind-spot and hit the right turn signal to get out of the passing lane, the car ahead abruptly stopped. That explains why the pink car drifted so slowly to the right - it wasn't an avoidance maneuver, it was a simple lane change while unaware of the stopped car ahead. Presumably an autopilot in the pink car would've been able to see the stopped traffic ahead and hit the brakes even while the driver was looking to the right to change lanes.
Same reason to use the handbrake at a crosswalk/zebra crossing: You don't want someone reckless behind you to make you cream some poor folks.
Safe follow distance isn't an evolutionarily-stable strategy. The number of accidents avoided by people following it is more than offset (at least psychologically in the minds of the drivers) by the perception that people are constantly cutting you off if you adhere to it. Since punishments for failing to adhere to it are few and far between, drivers get sloppy.
It's the sort of thing you could, however, program into a robot and expect the robot to adhere to it.
I've often wondered what people will initially do when being driven around by a robot. How anxious will it make them feel, not to give up control, but to see a robot driving "so slowly". Not going 10 over, letting other drivers in when possible, always slowing to keep a safe distance, not passing aggressively if someone is going a few miles under, etc.
In the far future, i suspect people will be so used to being driven that they won't care how fast they're going. They can browse, work, surf, do whatever they like. But the first wave of drivers being driven.. it's going to be interesting to see how people respond to it.
And that's not even accounting for the drivers who still driving, but seeing robots on the road. I bet they'll always want to pass robots asap.
No surprise, everyone hates them.
Here's an accurate article, which (alas) still has an inaccurate headline: https://electrek.co/2016/12/22/tesla-autopilot-speed-limits/
Which is perfectly fine - it's safer that way. Not wanting to feel like a 'loser' is understandable - but when it is life and limb that's on the line; I know which side I'd rather be. On the other side is road rage and stupid games of chicken with your life on the line.
Secondly, even if you are following a safe distance to the car in front and someone merges, causing a now less than safe distance, it is expected that you'll have to move back to a safe distance.
The only way that could be any different is if you somehow always kept 2-3 "safe distances" in front of you, and that way you were never at an unsafe distance if someone merges in front of you.. but that seems a bit of a slippery slope.
Regarding "how is it safe"... it's a safe following distance... i don't understand how you could even not consider it safe. What you seem to be suggesting is that "it's not safe enough, so i never want to be safe", which just seems a bit asinine. Ie, because if someone merges in front you now feel unsafe, you'd rather feel unsafe all the time by never leaving enough room for someone to merge and always driving at an unsafe distance.. that is just crazy, imo.
It likely is slower, you are correct, but safe driving isn't about shaving off 3 minutes from your morning commute. It's about safety, pure and simple.
If following distance < braking distance:
Brake
...We should just be allowed to carry police-verified dash cams which issues fines in live ;)
1) Approaching the SUV too closely (can't be sure of this one, it's possible that the SUV moved into the left lane just prior to the start of the video). Even at the beginning of the video, the hatchback was too close to the SUV.
2) Attempting a lane change while too close to the vehicle in front and the vehicle to the right. Lane changes are the most dangerous maneuvers in highway driving, because they divide your attention and may not be anticipated by other motorists. You want more space than you would normally need both in front and to the side you're merging toward.
3) Accelerating during to the lane change. I'm not sure that this is what happened; it's difficult to tell without stereoscopic depth perception. What I think happened is that the driver of the red hatchback accelerated during the lane change, approaching the speed they wanted once they were in the clear lane. This is very commonly done, but it trades away a lot of safety in exchange for a little time. In any significant traffic, you should complete the lane change first, then accelerate to the desired speed for the new lane.
I've seen this happen, thankfully without the crash and rollover, in my car (Model S). My car has stopped when the second car in front decelerated fast.
This kind of forward-looking speed control can also contribute to the melting of compression wave traffic jams.
Radar's a tough signal processing deal: a soda can on the road can mimic a large vehicle.
Possible next step in systemic forward-looking safety control, and much cheaper than radar: a few bits of data embedded in stoplights could announce the state of the stoplights of the next car forward (and even the one in front of that). Stoplights are all LEDs now, and they're bright enough to be seen by a really cheap camera even through mist on the glass.
This might not be a bad thing... in my early twenties I saw what I thought was a paper cup in the road and ignored it. When I hit it, my car felt like it was thrown up in the air - turned out to be someone's starter motor. I ended up with a cracked wheel rim and a flat tire.
wow - didn't realise this is a thing. Colour me impressed.
the system is indeed very interesting, and even the presentation is clear and useful. Serious kudos to everybody at Tesla.
One thing to note is that most dash cams, mine included, shoot at a fairly wide angle. The Tesla was likely following a good bit closer than it appears from the video alone.
This is also an interesting visual demonstration of why that Armco barrier between the opposing directions of traffic is so terrific. It's quite possible that having that in place saved several lives in this accident. It clearly absorbed some of the impact energy and prevented the SUV from crossing into oncoming traffic.
I wouldn't really want 'break' becoming an acceptable alternative, considering how cars braking and them actually breaking are likely to turn up in the same context and be really confusing.
(I'm not trying to accuse you of doing something wrong. This is just a minimal, relatively harmless example of a larger and more concerning phenomenon in this age of information spreading so quickly with so little verification. I suspect we're going to need to take a sharp turn towards self-censoring those little unsupported claims, among other strategies.)
(Personally, I see nothing wrong with posting anecdotal evidence as long as it's clearly identified as such.)
I currently drive a civic so am just guessing...
2: Minivans have an image problem (still associated with the "soccer mom" stereotype for Gen X; millenials appear to be less averse to them). In addition, most minivans are elevated so that they can have a flat floor, so they similarly have an elevated center of gravity.
Might be that millennial perception you're talking about, but I always thought soccer moms drive SUVs.
I would be interested to know how this compares with other vehicles if 'fatalities per mile driven' included fatalities that involved non-SUV occupants.
NB: I know nothing about this topic, genuinely curious to hear from someone who does.
They may have more structural material, but they're also heavier, meaning they have to absorb more energy.
1) heading straight while braking: probably the best case as you have the collapsible trunk between you and the obstacle, and you can reduce the speed, if by little;
2) spinning without control and crashing with the back or the sides: worse than 1), but not so bad as there are side airbags and the headrest. The car is at least somewhat protected on all sides, plus there's some tire friction for reducing the speed;
3) rolling over: less friction as the car is rolling, plus the car may crash into the obstacle with the car's roof or floor. Both are certainly more than a layer of tin, but not as protected. Plus, objects flying around the car and limbs in an unsafe position when the crash happens.
Small bump: http://www.ed.nl/polopoly_fs/1.6785912.1482854466!/image/ima...
Ouch: http://www.ed.nl/polopoly_fs/1.6785892.1482854220!/image/ima...
Source: http://www.ed.nl/regio/valkenswaard-e-o/valkenswaard/video-d...
Flipping also does not help stopping faster (in case there's something dangerous ahead). Instead, the faster way to break is by using the wheels just like the car is designed to do.
I remember I could easily outperform a couple of BMW SUVs on a highway in a descent from a mountain pass. They had passed me on the way up, more HPs and gravity helps handling. Then I kept turning at 130 km/h on the way down, where gravity is detrimental to handling. They were suddenly much slower, especially one of them (a prudent driver, I guess).
It can also be 20 km/h faster on wide straight roads but I still have to see that. I guess nobody is risking the bad press after the first accident at 150 km/h.
Did I mention I was surrounded by motorcycles?
On the very same road I wrote about, I've been driving another car that couldn't turn at 90 km/h uphill without drifting outside. It was not a SUV. Suspensions setup is very important for safety: a car that can turn fast is a safer car. I tested an Audi TT on ice at low speed (ice tires) and it was like turning on asphalt. Other more normal cars, not really.
The dangerous Toyota Hilux:
https://www.youtube.com/watch?v=xoHbn8-ROiQ
The Jeep Grand Cherokee acting similarly dangerously:
Interesting to see the Jeep blow the front left tire... 7 times. Consistent.
If Skoda has it I am sure that all other models in the VW Group's line (VW, Audi, Seat etc) should have it.
If you watch the video again carefully, the Tesla alerts before the car in front brakes, because the car in front of that is braking.
"Skoda Front Assistant is a system designed to help avoid or to mitigate accidents into the rear of preceding traffic. A long-range radar, positioned at the front of the car, can detect vehicles up to 80m ahead which the car is likely to hit unless action is taken."
Keep in mind, Skoda is not an upscale brand, they are on the budget end of things. The VW and Audi models might/should have better systems.
The new Peugeot 308 also offers a 'Emergency Collision Autonomous Braking System' (http://www.peugeot.co.uk/showroom/308/5-door/p=safety/) but on closer look this is actually 'an automatic brake application to reduce speed by a maximum of 12mph to reduce the severity of impact'. Much less useful, but at least moving in the right direction.
One problem with suddenly braking hard to avoid a collision in front is then you get rear ended (happened to me). Sometimes the best approach is to swerve onto the shoulder instead (also happened to me, and I watched the guy behind me plow into the guy in front).
This stuff is going to happen as the technology develops.
Detailed Explanation: In the UK part of the driving theory test is a "Hazard Perception"[1] exercise that test candidates awareness of hazards around them by playing short video clips and getting the candidates to click when they first spot a hazard they would need to respond to.
When watching the linked video I 'click' at 0:04 when I see the multiple brake lights though the car directly in-front. This coincidentally is when the Tesla responds with its audible warning.
The factors that lead to the Tesla not being involved in an accident in this video were not related to Auto Pilot but due to a competent driver: 1) Maintaining appropriate breaking distance from the car ahead to i. be able to stop in-time but without being tail-ended due to fast breaking ii. have 'thinking distance' to allow for slowed reaction timing 2) Watching the road ahead and noticing solid breaking of ahead vehicles though the directly in-front vehicle. There is NO AUTO PILOT MAGIC IN THIS VIDEO
I do not dispute that in other circumstances and perhaps other videos Tesla Auto Pilot HAS prevented and accident that a human would not have. This video is NOT such an example.
[1] https://www.gov.uk/theory-test/hazard-perception-test
Edit: Please see child comment https://news.ycombinator.com/item?id=13272626 clarifying my intentions with this post.
I think we'd all agree this isn't magic; the point is that it works in spite of the driver. A "competent driver" is going to be especially alert during an exam, but over the course of countless errands and commutes, their guard could be down, their reflexes may suffer due to lack of sleep, etc.
Autopilot doesn't get tired, and never lets its guard down. That's what matters.
My intention was in no way to underplay or diminish the achievement of Auto Pilot but to simply make the point that: In this video, of this event Auto Pilot adds nothing. In this case the driver was "alert and competent". This video does NOT showcase or demonstrate Auto Pilot preventing an accident, reacting faster or better than a human.
That is my point. People seem to be extolling how Auto Pilot saved that day in this video. It didn't.
Imagine if you could be confident that the cars sharing the road with you were all "alert and competent" at all times because they were using a solid autopilot system.
And yes, I would say that the autopilot reacted faster than a competent human driver not expecting a crash while watching a 30 second clip. Probably by time measured in seconds.
I do agree safe following distance made the auto-breaking pretty irrelevant in this specific case as there was plenty of time for a human to react. But it's still impressive to see things working, and a bit of positive public PR hype over this tech can't be an entirely bad thing.
You have highlighted a fundamental misunderstanding people have about Auto Pilot in its current form it is an SAE Level 2[1] driver assistance function.
It is not supposed to be functioning in spite of the driver. The driver is in charge and should be responding to events. It is dangerous to behave otherwise - examples include pretty much all Tesla Auto Pilot attributed fatalities.
[1] Level 2: The driver is obliged to detect objects and events and respond if the automated system fails to respond properly. The automated system executes accelerating, braking, and steering. The automated system can deactivate immediately upon takeover by the driver.
Edit: Formating
It's like saying cruise-control is worthless because any competent driver can maintain their speed - you're missing the point.
We're achieving parity between autopilots and human drivers, except the autopilot will never be distracted or tired, and always operates with the skill of a competent driver (and many drivers are not competent).
I can agree with this. Too bad it isn't the default sentiment instead of over-the-top optimism or cynicism.
I don't think anyone here is surprised that people get into stupid, easily avoidable car accidents all the time. The argument seems to be about where the technology currently sits. In this video the Tesla braked and avoided rear-ending the colliding cars; so did the vehicles with no autopilot in the right lane. This is "impressive" to some people, but it's also the bare-minimum level of acceptability for self-driving vehicles.
Not just a competent driver, but a competent driver that is primed to be looking for a hazard in a short video clip.
I doubt this assertion is true: lot's of cars with plain old cruise control can select a safe following distance and detect when the car ahead is braking. Roof-mounted LIDAR can see several cars ahead (and behind) and it would be negligent to to apply the same collision avoidance logic to car n+1
[1]: http://www.oecd-ilibrary.org/transport/road-safety-annual-re...
We get road fatalities per person and per vehicle from[2]
[2]: http://www.who.int/violence_injury_prevention/road_safety_st...
If we take them together:
i name p÷v *v/km
------------------------------------
1 Israel 0.371134 1.96701
2 United Kingdom 0.5686275 2.047059
3 Sweden 0.5957447 2.085106
4 Denmark 0.5223881 2.089552
5 Ireland 0.5394737 2.103947
6 Netherlands 0.5666667 2.55
7 Switzerland 0.7021277 3.019149
8 Germany 0.6323529 3.098529
9 Norway 0.7307692 3.215385
10 Iceland 0.7540984 3.544262
11 Australia 0.739726 3.846575
12 France 0.6710526 3.892105
13 Canada 0.6315789 3.915789
14 Austria 0.7605634 4.411268
15 Belgium 0.6261682 4.571028
16 New Zealand 0.7142857 4.785714
17 Slovenia 0.6736842 5.12
18 Finland 1.090909 5.236364
19 Spain 0.6981132 5.445283
20 Japan 0.7230769 5.784615
21 United States 0.8217054 5.834109
22 South Korea 0.4482759 8.158621
23 Czech Republic 0.7176471 9.975294
24 Malaysia 0.8026756 10.11371
25 Brazil 0.4652087 26.00517However, because I'm not British, I have no problem saying this:
You guys know how to drive, but if it makes you feel better, the Swedes are better.
Very often, humans do not do those things. And despite the carnage that results, over more than a century of experience with automobiles, we have not managed to make people significantly better drivers, nor come up with a licensing scheme that does more than weed out the most obviously medically-disqualified or incompetent people. At least here in the US, you take your drivers test (which is a ridiculous, comic farce) exactly once in your life, after which you can basically drive until you are blind, comatose, or--very rarely--you kill someone. But rarely the last one, not because people don't kill each other, but because it rarely results in a license suspension. We've just tacitly accepted that people are bad at driving.
So yeah, bring on the machines.
If you go by the street definition of driving competence, then you get exactly the accidents like the one on this video.
Safe driving distance. The primary thing most "competent" drivers don't give a fuck about. Second being speed limits.
And yes, I understand many people drive that way not because they want to, but because other drivers aren't leaving them a choice. Ironically, this area is already heavily regulated. What I believe is needed is much, much stronger enforcement of those regulations.
With some software I'm sure they can determine some unavoidable accidents before they happen.
What I'm curious about is whether the autopilot predicted the crash due to the relative speed of the two cars ahead of it, or if it was reacting solely to the change in speed of the car that braked.
I guess the point is to strike a balance between trust in the automatic systems and trust in the driver to make the right decisions.
Another way to look at it is that the system is designed to protect you, not the car. Use automatic braking to bring the crash down to something the car can protect you from, the rest is up to you
Another is saying: This seems like a very convenient piece of good PR for Tesla (and self driving car technology)...
Yet another is saying: I hope the folks in the car are alright
Why can't we appreciate tech advances without devaluing people?
"Calculator" used to be a job performed exclusively by humans. Moving to mechanical & digital calculators didn't only require gains in efficiency, but also a recognition that this was a job humans were simply worse at than machines, so it was irresponsible to task a human with it.
Similarly self-driving cars can't simply be a nice thing to have, eventually we have to have the discussion that it's irresponsible to have human driven vehicles on the road.
I'm not arguing that we should endanger people and my comment is not about this one instance. It's about a general attitude of devaluing people. Where do you think that road leads?
When Uber vehicles were found to have run red lights a couple of weeks ago, Uber's response was that these were human operated and served as an example of why they must rid the roads of error-prone human drivers as quickly as possible. This, as if humans were a scourge that must be eliminated. There is a strange rising anger with people for being human.
And, once we've ridded the roads, labor force, economy, etc. of all the pesky humans, then what? Who owns the tech? Who has the power? Will they be as benevolent as the machines we've learned to worship? What is the value of a human life by then?
Quote whatever stats you like, but normalizing this attitude of fallible, dispensible humans is dangerous in many ways.
Is pointing this out somehow wrong? I think not pointing it out devalues human life.
The road of getting rid of fallible humans from boring automatable jobs leads to Utopia, and I reject your notion that this is somehow inherently different from e.g automation in agriculture or computing
I've been clear about my meaning. Can't simplify it any further.
>Because certain types of technical advancement requires an element of devaluing people
That's your quote. What did you mean?
>Of course a more error prone, dangerous and expensive driver has less value.
There's a difference between preferring tech to do the driving and saying the human has "less value". You don't see that?
In any case, a hostile regard for human fallibility that approaches promotion of the dispensability of humans isn't a good thing.
That's not really what I had in mind with my comments here, but it is an interesting point.
And maybe it is another reason to consider prudence as we rush headlong into a world increasingly reliant upon technology.
We are approaching a very scary time when this conjecture will be subject to cold empirical evaluation.
I think we'll pass the test. But wishing that humans were good on things we aren't do not help anybody.
You really haven't. This entire thread is you taking disposablezero's "Relying on error-prone, distracted, slow humans to guide massive killing machines is playing Russian roulette" comment out of context as some moral statement about human beings.
There's a bunch of things humans are worse at than machines, you're the only one asserting that this somehow "devalues" people without any further clarification.
I'm also worse at shoving nails into wood than a hammer is. How does that devalue me?
> [Because certain types of technical advancement requires an element of devaluing people]. That's your quote. What did you mean?
The set of things machines are better at than humans is only increasing over time. Today nobody disputes that a combine harvester is better for the task than a team of humans, but there's a tendency to overvalue human work as machines are getting better.
I've noticed that changing with "handmade" within my lifetime, and we're probably about to see it change with "human-driven vehicle" within the next decade.
> There's a difference between preferring tech to do the driving and saying the human has "less value". You don't see that?
No, I really don't. You haven't defined what you're talking about when you're saying "value" or "devalue", but I will. I'm talking about value in the economic sense. I.e. if you've made a combine harvester instead of needing 50 people to do manual labor for 16 hours a day you've optimized your economy and added net value to our civilization.
I really don't buy this argument that saying a human has less value is a negative. The logical conclusion to that argument is saying that we should unwind civilization and technological advancements until we're all hunter-gatherers again. Because surely the existence of the food industry devalues hunters everywhere.
> In any case, a hostile regard for human fallibility that approaches promotion of the dispensability of humans isn't a good thing.
No, it's the exact opposite. It's saying that we shouldn't waste a human being's potential on some menial task like manually picking corn in a field or driving a car.
To someone living in the future we're hopefully heading towards the notion of someone needing to make driving a vehicle their full-time job will be as ridiculous as the notion of of a human being needing to manually carry freight is today, as opposed to using a truck.
Well, I guess I haven't. Perhaps that's because there are really a myriad of contexts for the word "value" here. For instance, you've chosen to focus mostly on "economic value".
And, sure, I have an argument with that too. For instance, you say this:
>It's saying that we shouldn't waste a human being's potential on some menial task like manually picking corn in a field or driving a car.
Sounds great, but nothing in recent history bears out a desire for this as a driving force or even a significant consideration. In fact, the reality is so far off that this argument is intellectually dishonest. That is, we're not handing over millions of people to a life filled with purpose, now that they are released from the "menial tasks" that were once destroying their potential. Instead, we're really just leaving them rudderless (and penniless) in a society where our value as humans is significantly defined by our work/economic-output.
Further, the economic benefit of this automation accrues to a relative few at the expense of those "newly freed" individuals. It is dramatically redistributing power and wealth and threatening self-determination as well as the democratic model. In short, it tacitly promotes a very real form of oppression.
So, piling onto that with a further denigration of the value of those being disenfranchised is unhelpful to maintaining a healthy society--to say the least.
>The logical conclusion to that argument is saying that we should unwind civilization and technological advancements until we're all hunter-gatherers again.
You have to know that's classic reductio ad absurdum. I've never advocated that we don't leverage tech. But, if you really want to carry an argument to a logical conclusion, consider the very real eventuality that all jobs will be displaced by technology, including "knowledge-workers". That's the difference between the current reality and the relatively low-tech technological displacements of the past to which you keep alluding: here, even skilled labor is at-risk.
This is happening on a continuum. So, I'm suggesting that it is becoming increasingly more important to define our relationship to technology in healthy ways.
Instead, what I'm noticing overall is that there is a certain hostility towards humans (devaluation) when championing tech these days. Whether it's in the labor force, with autonomous vehicles, or otherwise. Until more recently, we've known about and relied upon various tech to keep us safer and perform other critical functions. But, they played those roles as enhancements to humans. Increasingly, however, we speak overtly about replacing humans and we juxtapose the value (any value) of a human with that of technology. It's a dangerous framing.
But, it seems we just disagree here. You think there's nothing wrong with that. I do.
Indeed. If only we could find them all and kill them before they strike again.
Either way there is a huge difference between "I'm fine with hurting people's feelings" and "I'm fine with killing people"
Those aren't the choices.
The romanticism of "human control" is not an idea I am willing to support. Humans are terrible at most things they do. We get tired. We get distracted. We become complacent. We bend the rules. We forget steps in the safety protocol. We make incorrect decisions. If there are only 99 ways to do something wrong a human will invent the 100th way of doing it wrong.
If mechanical failures and software errors can be proven to be a safer risk than human error - then humans should be exchanged for robots. Jobs and hobbies be damned if there are lives at stake.
But that should be pretty straightforward; there's a lot of things people can do that aren't "drive a car."
I'm a lot less inclined to take that attitude in a space where there were 35,000 deaths due to vehicular accident and the most common causes of accidents are distracted driving, drunk driving, speeding, and reckless driving. That's basically "humans," "humans," "humans," and "humans;" I suppose we could spin that by pointing out all the millions of miles per year people drive successfully without killing themselves or each other, but that doesn't change the fact that removing humans from the equation should push the death numbers down further.
Driving is at that special saddle-point of attention-thirsty and tedious that people can be good at, but not perfect; the models of how attention wanes in a tedious environment are pretty solid. It's precisely the sort of thing we should be trying to automate if we want to drive the fatality rate below the current 11 per 100,000 population number in the US.
That's stating the obvious, not devaluing people.
In any case, your argument would be extremely effective--if my argument was that people are infallible.
Why does every single bit of Tesla "news" make the front page? I see 50 videos a day on Reddit of people miraculously avoiding accidents, never once on HackerNews. This is kind of cool, but I'm going to remain skeptical that it "predicted" the accident. There were other cars that also stopped in time. Did we expect it to bulldoze into the back of the cars?
I mean, do people actually believe that Tesla has developed self-driving AI that is actually "predicting" accidents? Get enough cars on the road using the technology and we'll see amazing things that are hard to explain.
I don't understand what would be difficult to believe about this. It doesn't take super-sophisticated math to predict car X and car Y are too close and moving too fast to avoid an accident. It's mostly a matter of having the sensors to build an accurate map, and it seems like Tesla has that.
It seems like it could instantly react to the car in front drastically changing speeds (read: colliding with the car in front of it), and brake immediately. But predicting seems like a different issue, with far more false positives (Tesla hitting the brakes when it anticipates a potential collision). Is that happening?
You could be right, but this video certainly hasn't convinced me.
Don't understand the downvote. It was a simple question from someone who is concerned about (online) privacy.
" In such situations, you should pray that there's no truck behind you." -- So true. Wouldn't the autopilot account for that before braking that fast? It would cause a much bigger problem if not...
in my country the police does not even bother with you till 160 kmh. (130 is legal limit)
That said, 100km/h made sense in 1970, but now with the safety of modern cars, there's no reason that 130km/h shouldn't be the standard everywhere modern cars are prevalent.
as you can see 130kmh is used by most countryes in europe
Not the brake light: the speed differential between that car, the car one car ahead, and the car with the radar.
This is the bit driving schools rarely (if ever?) teach and which many drivers never get a chance to think about until they find themselves in a stressful situation where they can end up making dangerous decisions. (And then, of course, most people do not think).
Fortunately, they were right behind the accident. If the wreckage was stopped and partially blocking the left lane, Tesla's autopilot probably would have plowed into it. Like this fatal crash in China[2] and this non-fatal crash in Germany.[3]
[1] http://www.roadandtrack.com/car-culture/videos/a32073/no-tes... [2] https://www.youtube.com/watch?v=fc0yYJ8-Dyo [3] https://www.youtube.com/watch?v=TqfgDrynm78
Also Tesla's radar which can detect vehicles 2 cars in front of you by bouncing signals off the car in front is pretty neat tech. Who else is doing this? I'd assume most of the luxury car makers with auto pilots have this too?
coordinating this stuff doesn't require computer vision or truly self-driving cars or anything like it and could have been done 30 years ago (1986) over an AM radio standard and some kind of coordination between cars - just nobody sat down and designed that standard.
today wireless coordination technology offers dozens of choices (in the gigahertz domain) and caravans could be assembled all but "trivially", cars can know where they are via gps with no problems at all. That would be an exciting move forward. Of course, it only works when everyone is doing it, but if just a few traffic jams instantly disappear (the ones where are the cars happen to have caravan functionality and coordinate) it would improve things for everyone: the traffic jam will disappear whenever it bunches up in a way that happens to consist of coordinated cars. (e.g. a 3-car phantom jam slowly moving backward will disappear whenever it crosses 3 caravan-enabled cars - not every car has to have this functionality). But I haven't read anything about Tesla coordinating even with other Teslas, let alone some standard of intercar coordination. A shame - this stuff is way easier than the self-driving stuff Tesla and others are doing. It's very low-hanging fruit.
There is this "V2V" initiative by NHTSA:
https://www.nhtsa.gov/technology-innovation/vehicle-vehicle-...
How do you direct the signal to only cars in your lane? In your direction of travel? How much efficiency is lost if all of the drivers don't give themselves enough space between the car in front of them and their own vehicle? Now they have to wait to move forward. How do you get all drivers to roll forward together at the same rate of acceleration?
You could have the last few questions handled by a computer, but you'd still have to overcome a lot of other hurdles (namely location awareness, peer-to-peer communication, infrastructure changes to road signals if you don't have self-driving cars which can detect light states, etc) before you could even accomplish something this simple at scale.
it's trivial. To show you this is trivial without pretending to be an automotive engineer I'll use the example of a screaming auctioneer who sits on top of your steering wheel. A conversation with him might go: "HEY I NOTICE WE'RE STUCK IN A JAM IS IT OKAY OF I SIT ON YOUR STEERING WHEEL AND OBSTRUCT YOUR VIEW AND THEN I'LL START PRESSING DOWN ON THE GAS TOGETHER WITH ALL THE OTHER CARS IN THIS TRAFFIC JAM SO WE ALL START MOVING AT ONCE OKAY OKAY OKAY JUST TELL ME IF YOU'VE HAD ENOUGH READY OH LOOK WE HAVE CONTACT BETWEEN THE BUMPER TRANSMISSION LED AND RECEIVER LED AND THE CAR AHEAD OF US'S REAR BUMPER TRANSMISSION LED AND RECEIVER LED OH THIS IS AWESOME THE CAR IN FRONT OF US ISNT'T THE ONLY ONE WITH THIS EQUIPMENT THE CAR IN FRONT OF IT ALSO HAS THIS AND LOOK THE CAR IN FRONT OF OF THAT ONE ALSO HAS IT SO WE'RE ALL HOOKED UP AND AND WHEN THE FRONT CAR STARTS MOVING WE'LL START GOING AHEAD TOO ALL RIGHT JUST PRESS THE BRAKE IF YOU WANT OUT OF THIS BECAUSE WHEN THE FRONT CAR STARTS MOVING I'LL PRESS THE GAS AND MOVE TOO! HEY IT'S MOVING!!! I'M PRESSING THE GAS!!! PRESS THE BRAKE IF YOU WANT THIS TO STOP!!!!! WE'RE MOVING!!! WE'RE MOVING!!!!"
it's an absurd example but shows that it's trivial. The only thing necessary to get a stalled phantom traffic jam moving is for all cars to slowly start moving together. they can maintain distance between each other.
it's trivial if it requires special equipment such as bumpers that communicate with the car in front or behind them, proximity sensors/detectors, a way to transmit this information intercar and a way for the system to collectively brake or apply gas (like cruise control).
if you're an engineer I'm sorry if you can't see how absolutely trivial it is COMPARED WITH SELF-DRIVING CARS. It could easily have been done in the 80's with zero environmental perception, zero awareness of road conditions, nothing, and not hooked up to the steering wheel in any way. You don't need the system to handle any steering, period, in order to be able to break up phantom traffic jams.
if you've never invented anything in your life then you'll have to take my word for it that this is trivial, sorry. We're 30 years behind schedule on it and have suffered 30 years of phantom traffic jams that could have disappeared whenever they made their way to high-end cars that could have been equipped with a limited caravan system to start moving together with the drivers' permission. A phantom traffic jam can disappear at every point. Even if just 5% of cars were equipped with the standard.
you don't get what a huge deal it is for a few cars to be able to start moving together, coordinated, instead of one after the other. You don't get how 500 ms - 2500 ms of human driver latency means that traffic jams form which otherwise wouldn't and that once it is formed it is impossible to brake up and snakes its way back - but could be broken at ANY point by any group of cars that briefly acted as a caravan. you'll have to just trust me that the technology for this is easily in the realm of what could be done in the 80s. self-driving cars are lightyears ahead of that.
I would not be surprised if Tesla were subpoenaed (or otherwise requested of) for the data, in a case like this (similar to the recent Amazon Echo news, or cellular providers with geolocation in abduction cases).
Does anyone know how I can get my hands on a nvidia drive PX2 board and a quanergy lidar? It seems the lidars and such nvidia boards are only sold in huge quantities to auto makers and not as single units available for people to make their own robots.
I don't believe the car is clairvoyant, but my guess is radar bouncing off the asphalt under the cars. And maybe a glimpse of the brakelights?
I live in Dubai and it is a mix of many cultures and people drive like crazy here, I have seen a single accident involving 6 cars because most of the rash drivers will drive tail-gating
At least Dubai really needs lot of Tesla cars to avoid such accidents which are very common here
The Tesla would be aware of the cars behind it as well, so this sounds unlikely
Erm (not a radar but cameras + CV/ML)
/s
The Tesla vehicles use cameras, ultrasonics and radar. So they are processing graphical data (a video stream) and signal data. That's exactly the kind of task that GPUs are good at.
However, this only gets you as far as perceiving the world. The vehicles also have to decide which action to take -- following the traffic laws of the jurisdiction they are in -- which is exactly the kind of branchy imperative task that CPUs are good at and GPUs are terrible at.
Yes, that's it. Signal processing for forward looking radar, forward and backward looking cameras, and side-looking sonar sensors. Definitely local.
Cloud wouldn't work. The Teslas have LTE data links, but their latency is grossly unpredictable.
https://blogs.nvidia.com/blog/2016/10/20/tesla-motors-self-d...
Not sure what HW their previous cars had
It is easy enough to find it though: https://www.youtube.com/watch?v=D4xt2smoUso
And to support your comment - yes, this isn't exclusive to Tesla by any means!