Watch a Tesla Fail an Automatic Braking Test and Vaporize a Robot Pedestrian | Hacker News Reader

Watch a Tesla Fail an Automatic Braking Test and Vaporize a Robot Pedestrian

thedrive.com

66 pointstomohawk111 comments

The IIHS does all sorts of autonomous braking test scenarios involving pedestrians - https://youtu.be/TJgUiZgX5rE

A pedestrian darting out from behind a car is known to be a tricky scenario, but some systems handle it at least some of the time. This scenario looked like the pedestrian was unobstructed and traveling in a predictable direction, so it’s pretty unnerving if it’s real.

Long story short, even in Teslas, even just autonomous braking and crash avoidance is a great extra safety layer, but far from something that should be relied upon. Tesla is reckless for selling it as something that can be relied upon to successfully pilot a car.

This is a common issue across all vehicles with radar AEB systems, and the problem isn't vehicle manufacturers but the general public in the belief that these systems are more robust than they are (pedestrian collision avoidance).

To your point about Tesla, they vigorously communicate that you (as do other manufacturers) as the driver are responsible at all times. We can go roundabout like we always do about this, but regulators have continued to allow these systems to be marketed and sold, and they're the authority. Why are we blaming automakers for garbage drivers? The driver is who you sue when they hit a pedestrian.

> To understand the strengths and weaknesses of these systems and how they differ, we piloted a Cadillac CT6, a Subaru Impreza, a Tesla Model S, and a Toyota Camry through four tests at FT Techno of America's Fowlerville, Michigan, proving ground. The balloon car is built like a bounce house but with the radar reflectivity of a real car, along with a five-figure price and a Volkswagen wrapper. For the tests with a moving target, a heavy-duty pickup tows the balloon car on 42-foot rails, which allow it to slide forward after impact.

> The car companies don't hide the fact that today's AEB systems have blind spots. It's all there in the owner's manuals, typically covered by both an all-encompassing legal disclaimer and explicit examples of why the systems might fail to intervene. For instance, the Camry's AEB system may not work when you're driving on a hill. It might not spot vehicles with high ground clearance or those with low rear ends. It may not work if a wiper blade blocks the camera. Toyota says the system could also fail if the vehicle is wobbling, whatever that means. It may not function when the sun shines directly on the vehicle ahead or into the camera mounted near the rearview mirror.

> The edge cases cover the gamut from common to complex. Volvo's owner's manuals outline a target-switching problem for adaptive cruise control (ACC), the convenience feature that relies on the same sensors as AEB. In these scenarios, a vehicle just ahead of the Volvo takes an exit or makes a lane change to reveal a stationary vehicle in the Volvo's path. If traveling above 20 mph, the Volvo will not decelerate, according to its maker. We replicated that scenario for AEB testing, with a lead vehicle making a late lane change as it closed in on the parked balloon car. No car in our test could avoid a collision beyond 30 mph, and as we neared that upper limit, the Tesla and the Subaru provided no warning or braking.

https://www.caranddriver.com/features/a24511826/safety-featu...

> Why are we blaming automakers for garbage drivers?

Because most people are bad drivers, most of those people realize it whether they'll admit it or not, and I truly believe that deep down most people recognize that they are one inattentive moment away from killing someone.

There is a desperate desire to be able to offload that responsibility onto someone or something else. "They came out from behind that car" instead of "I shouldn't have been going 45mph next to a row of parked cars." "They stopped short" instead of "I was tailgating."

If we can offload the responsibility it becomes someone else's problem and we don't need to change our habits of driving too fast, not paying enough attention to the road, and being too distracted.

I really have to push back against your point that Tesla “vigorously communicates” that driver attention must be maintained at all times, when in fact, they really seem like they want to occupy a gray space in-between:

- They advertise and sell full autonomy under the name “autopilot”, which carries all sorts of connotations

- Correct me if I’m wrong, but they don’t actually prevent a driver from falling asleep behind the wheel, even if they vibrate the steering wheel. New Subarus have driver eye tracking built in - it would be trivial for Tesla to make sure the driver is paying attention, but they don’t really want to.

It might say it somewhere in the manual to cover them legally, but all signs point to them wanting the market to think their cars drive themselves.

You're pushing back against the wrong part of my comment. You don't believe Tesla is meeting some sort of bar (re: your reference to Subaru), but regulators have approved of Tesla's continued approach regarding driver awareness (periodic steering wheel torque dead man switch versus requiring eye tracking). So what evidence are you using that your approach should be required and would be superior? Drowsy and intoxicated people still drive every day, killing innocent people, without driver assist systems (roughly 100k accidents a year are from drowsy driving, and about 10k people a year are killed due to drunk drivers).

You cannot mitigate away personal responsibility failures with technology bandaids. At some point, the buck must stop with the operator. (On my large shop equipment I let others use in my workshop, there are large stickers that very loudly proclaim "Not only will this kill you, it will hurt the whole time you are dying").

I don’t think regulators are actually doing their jobs here. If Teslas actually a successful dead man switch, we wouldn’t see story after story like this -

https://amp.theguardian.com/world/2020/sep/17/canada-tesla-d...

I don’t need to put forth a solution to point out that Tesla is trying to have its cake and eat it too. They say one thing to regulators and another to consumers and they are being allowed to do it.

I don’t disagree with you about personal responsibility, but Tesla also has responsibility for what they market and the downstream effects.

We will agree to disagree, but I still appreciate the conversation. I encourage you to write a letter to the regulator of your jurisdiction if you feel your concerns aren't being addressed. I have done the same with US regulators (NHTSA) in support of Tesla's driver assist systems.

I've driven a car with lane keep that worked like this:

1. If the lane keep is on and there is no torque warn the driver. If the driver doesn't respond, get really aggressive, pulse the brakes and eventually stop in the middle of the road.

2. If lane keep turns off for any reason (for example, momentarily too light lines) all driver monitoring immediately ceases.

Noone writes articles about this system that is not Tesla, but I find it impossible to believe that it is a safer system than Tesla's.

You have to maintain a torque pressure on the steering wheel. Otherwise it will alert and ultimately bring the car to a stop with the hazard lights on and disable autopilot for the duration of the trip. The system notifies you that you must keep your hands on the wheel and keep your attention on the road every time you activate it. It's communicated constantly through text, flashing color changes on the screen, and audio signals.

Nissan performed quite well in those tests, I'm impressed.

Many of the comments here seem to be missing the point. It is an object moving into the path of the car, and if the safety systems work as advertised, it should avoid that object. Whether it's a person, shopping cart, bike, go kart, etc. doesn't matter.

Exactly - that object could have just as easily posed a mortal threat to the driver of the Tesla, yet it didn’t stop.

It's also worth saying that the human brain can almost instantly do a fight or flight calculation as to whether it's a moose or a plastic bag floating by taking into account the motion, and it's reflective properties etc.

This task clearly is very very hard to do, but tough shit: if you can't safely control a megajoule of kinetic energy coming down the road your computer shouldn't be given the benefit of the doubt.

This test is so poorly done it feels like it must have been done for propaganda or bad PR.

Aside from the legs barely moving, the arms not moving - e.g. it doesn't look like a pedestrian the AI would be modeling from - it's driving out of context through parking lot lines.

Edit to add: the guidelines for using Autopilot, irrespective of what the name insinuates to you, requires the driver to be paying attention at all times and takeover if it's not doing what you expect; "BUt iT shOUld BE PerFeCT" attitude is dismissing one's personal responsibility - if you don't want to have to pay attention then don't drive with Autopilot.

It could be a 100-ton concrete block in the road. You don't want your car to drive in to it regardless, no matter if it's a person or not, and no matter what the lines on the road look like. You car driving itself in to anything is bad.

"Sorry you lost a leg, but to be fair, you don't walk like the type of people in our training set"

This is sort where jaywalking originates from i.e. people are getting run over so make it illegal to cross

Should it matter? It's still an obstacle that it shouldn't be hitting

Even if it doesn't move its arms like a prototypical walking human, shouldn't the car still stop, rather than ram into an unidentified large obstacle / skateboarder / pedestrian carrying something?

If it's driving in a parking lot rather than a busy city street, shouldn't that have meant it can also safely swerve away from the obstacle in almost any direction?

As others have pointed out, the arms not moving shouldn't prevent the vehicle from braking.

There are a number of reasons a person's arms may not move as expected, a couple of primary ones being the person is holding an object or has hands in pockets. And of course, some people don't have arms, and they shouldn't be left out either.

>This test is so poorly done it feels like it must have been done for propaganda or bad PR.

I disagree. The test is not poorly done. Such conditions are trivial for a human to recognize. You will not be allowed behind a wheel if you could not avoid a dummy while driving.

If an AI can't recognize such trivial circumstances why would I let it take control when I'm going 60 mph on a freeway? That's like putting my kid on my lap and letting him drive... yeah of course I can take over control at anytime, but it's still illegal to do this and absolutely insane.

Let's put it this way, if it's insane and illegal to put a 7 year old kid on your lap and allow him to drive at 60 mph down a freeway (even when you have executive override abilities to take control at anytime) then why is it normal to have an AI (who would do considerably worse than a 7 year old at recognizing a pedestrian) do what's basically the same thing?

Is this Bad PR or is it a Tesla owner trying to justify his purchase?

"the guidelines for using Autopilot, irrespective of what the name insinuates to you, requires the driver to be paying attention at all times and takeover if it's not doing what you expect"

This is similar to the FAA's "see and avoid" rule. It is mostly ineffective for obvious reasons, but is a good way to dump responsibility on one of the entities involved.

I can't quite tell if this post is a parody of the kind of comments that turn up on every single Tesla article, or if it is actually one of the kind of comments that turn up on every single Tesla article.

Even if you totally ignore the fact it's person-shaped, a self-driving car should still stop if it's going to drive in to something.

Imagine you are on the highway in self-driving mode and there's a plastic bag flapping around on the road ahead.

Probably best NOT to stop in that scenario.

It's the millions of "edge cases" that humans are pretty decent about making quick decisions on that make self-driving cars really scary to me.

Probably best NOT to stop in that scenario.

The car should slow down for that scenario. It doesn't know if the situation is 100% safe, so it should apply the brakes and steer to avoid a collision using data from around the car to determine the safest course. Driving through a hazard is only an option for a self-driving car if there is literally no other option.

Cars are not people and they're not capable of making decisions in the same way people do. They should be taking the safest route all the time.

It should not stop in that scenario because there is no hazard. Wild application of brakes is just as dangerous as not stopping in a hurry when necessary.

I've not driven a Tesla, but I'm going to guess they have different levels of braking between "not braking" and "wild application of brakes", so when I said it should "slow down" I obviously meant something in the middle.

Most people would describe the car suddenly coming to a halt apropos of nothing to be "wild application of brakes". Cmon.

Driving through a plastic bag also just isn't a safe option in rare cases for cars. It can get wrapped around something and cause some damage or latch onto the exhaust and burn and smell awful.

Do people here really just drive through road hazards without being cautious?

If you brake in the middle of the highway (even just to slow a little) when the road ahead is clear you become a bigger hazard than the plastic bag. Double if you swerve as you suggested.

The road ahead isn't clear though. A self-driving car knows something is there but it doesn't know what it is or what it might mean about potential future events. You and I know a plastic bag is likely to be a one-off non-dangerous thing but a car doesn't (it might know about a person, or a bag, or another car, but it won't recognise all of the things it could be). Consequently it has to be cautious and should take action within the bounds of what is safe using data from around the car.

What if its a pidgin? You want to smash into that at full speed?

Many driver-assist programs ignore distant stationary objects because the sensors aren't good enough to tell if they're really in the road or just close to the road, or maybe a metal plate that should be driven over, or whatever. But the collision detection for close-up objects is usually better than this.

One time years ago, I was driving on a four-lane highway when I observed a plastic bag flapping around the road ahead of me. Other cars were slowing or swerving to avoid it, which I thought was peculiar, so I also slowed a little bit but (admittedly) mostly to avoid hitting the other cars.

As I passed, I learned that the plastic bag was actually a dog that was (for unknown reasons) trying to cross the highway in moderately heavy traffic. It was pure luck on my part that I didn't hit the dog as it ran right behind my vehicle.

Sometimes even humans aren't the best judges in a split second of an object's estimated mass and/or whether it might contain some sort of soul. (I was around 30, and neither my cognition nor vision were noticeably impaired)

Similarly, I was on a highway and saw what looked like a crumpled up brown paper bag in the center of my lane and, rather than make the large maneuver of going around (I had passengers and didn't want to jostle them), I chose to straddle it. Welp, turns out it was a brown rock that had fallen onto the roadway. It managed to clear my oil pan (yay!), but didn't clear the floorboard and rotated longwise and propelled a bottle that was in a grocery bag on the floor into the face of one of my passengers.

I dislike Tesla's misleading FSD claims, but this is a pretty unfair test imo. They timed the movement of the pedestrian so it was only in front of the car when it had less than 10 feet to stop, of course it will hit it. I don't think we should have the expectation that automatic braking is always able to predict someone's movements, regardless of how good your system is cars need more than 10 feet to stop. And the car actually did stop, it just couldn't stop before hitting the pedestrian because it had no distance.

Edit: the second test is pretty bad for Tesla though, the dummy was stationary

You and I see very different things when we watch that video. There's absolutely no doubt in my mind that a competent human driver would have avoided hitting the dummy.

The dummy in no way looks human in the context I’d expect a self driving system to have. Humans don’t have a way to move their legs back and forth that way while remaining stationary (unless on a treadmill I suppose).

You can infer that it is meant to represent a human, and thus likely to move across a crossing, but there’s no reason to expect a computer to read it as a human right?

The most obvious characteristic is distinctly not human (Legs don’t work that way), so it looks like some weird windmill, or one of those hot air moving men.

I wouldn’t my car to drive into a wheeled mannequin rolling onto the street either.

Do you think it's fine to hit things that don't correctly get categorized as human? Like maybe someone on crutches?

Yes, this sounds straight out of a dystopian future novel. Missing an arm or leg? You will be hit. Extremely short or tall? Hit Walk with a limp? Hit Pink hair? Hit

Why does it matter though? Its an obstacle. When would you want to hit and obstacle anyhow? (Ok I can think of few but, no need for dark humour)

1) I'm not joking. I think the OP legitimately didn't think about the humans that don't look like they expect them to look. That's a huge problem across all of tech - not just our lack of imagination about the wide array of human experience, but us assuming that we have thought of everyone when we clearly haven't

2) I think this part gets trickier How do you tell an AI not to do irreversible things? How do you indicate that this is a plastic bag blowing across the street and it needn't cause a traffic jam, but that is a plastic tarp hanging off a barricade that we can't drive through?

How do I indicate that this is a piece of cardboard we can drive over but that is a human person wearing a Halloween costume made of cardboard?

It's not an easy task, and I'm very sympathetic to the very real technical challenge they're facing

I'd expect a car with an automatic braking system to automatically brake when a "weird windmill" or "one of those hot air moving men" moves predictably to intersect its path.

Yeah true, only cars and humans are allowed to be on roads.

Everything else is magically repelled away. Like debris, crashed cars that don't look like cars, animals, stuff that fell of that truck, a soccer ball kicked from nearby field...

Your attitude seems pretty bad.

I don’t think that was the point I was making at all.

This seems exactly like a case where computer algorithms would function worse than humans. It is not surprising (to me) that those exist.

For instance, put a sign that flashes “Skydivers will land 100m in front of you in about 5s, break now” and human beings will have 100% success rate in avoiding it, and computers will not. (Or any other situation that relies on written communication that is not usually used on the road).

There’s plenty of situations where human ability will perform better than computer models (and vice versa).

I think you are being too generous with what a self driving car should be able to do.

If you were moving a house and was caring a big box across a street. Its not unreasonable to expect that you will be able safely make it across the street, even though you don't look exactly like a normal pedestrian.

And it is not unreasonable to expect a self driving car to break if there is any type of a obstacle in their path, especially if not classified. We are not talking about some wacky scenario. A bird might be flying across a street at windshield level. A ball might be kicked onto the street. A hey-bale might falloff onto the street from a tractor driving ahead. A huge box with legs is crossing a street.

I am not expecting a child to be hiding behind a parked car jumping out in front of incoming car to be a scenario that anyone could catch.

My question is why those car dont break when there is object infront of the?

I don't care if it looks like a box or a goat. I'd expect an Automatic Breaking System to not hit it.

Throw a dummy in a powered wheelchair. Surely the car isn't biased against wheelchairs. Or put a pendulating dummy in crutches. Or one that walks with a Duchenne Gait (as in muscular dystrophy) There are dozens of situations where humans don't "walk like humans" and your bias is showing if you only account for "normal" gaits.

The second video had audible warnings of an imminent collision. Continuing to apply the accelerate is effectively an override.

In Chinese language video, they explain that they are testing for a specific kind of accident, a “Guitantou,” which is when a pedestrian walks in front of a car with no warning. The video isn’t trying to be dishonest about what they’re testing, it’s just that whoever wrote this article didn’t bother to check what the announcer was saying.

While I agree that the test is unfair (because of breaking capabilities), this is by no means a crazy situation that wouldn’t happen in a city or a place with tons of children. The important question this rises, and the most important question regarding self-driving cars in general is: who is liable?

Until that question is settled in court I don’t think I’d feel comfortable using the self-driving capabilities of a Tesla.

It's not unfair but realistic!

No one expects all cars to always perfectly breake in all situations.

But Tesla breakes far from perfectly.

And this isn't the first time Tesla has worse then everyone else results in emergency breaking system tests.

The driver is liable. You wouldn't use cruise control on city streets with pedestrians around. You shouldn't use the self-driving features either.

If we're really talking about automated driving where humans are only passengers, that's an entirely different kettle of fish.

How is this test unfair? Any human would have swerved or at least braked.

This is not a difficult scenario.

Unit testing don't check for the obvious.

They timed the movement of the pedestrian so it was only in front of the car when it had less than 10 feet to stop, of course it will hit it.

Here in the UK part of the driving law is that you should always be able to stop in time to avoid hitting something. If you're driving somewhere with poor visibility (eg a road with cars parked down the sides) then you should slow down. If you do hit something it's determined that you were driving too fast, so it's considered your fault and you have to have evidence to prove otherwise (eg a dashcam).

In the case in the video, seeing someone waiting by the side of the road should be enough to make a competent driver slow down a little.

I imagine the same is true in the US and China.

That was my first reaction — why is the car not slowing down at the first sight of something that looks like a human at that distance? Might be perspective but it looks like it's driving far too fast to begin with, not to mention to avoid a collision.

If Tesla's have such tunnel vision that they can't react to someone clearly walking into their path, then that capability shouldn't be allowed on the roads. A human driver would be 100% at fault in that situation, self-driving car manufacturers shouldn't get leniency because it's hard to not kill people.

Is it unfair to have a pedestrian dart out onto the roadway? That is a very predictable occurrence, any FSD developer should have pedestrians darting out as the top of their priorities list. And it's way more important to not hit pedestrians than other cars, they have so little defense.

Honest question, if the driver accelerates does it overdrive the AI?

In other words, if the autopilot would break, but driver accelerated, what would the car do? Also, what SHOULD the car do?

I'm asking this because we have no idea how the tests were done: we're given a video of a moving Tesla hitting a dummy and breaking after.

I don't know how people can take any conclusions from this information, other than what is shown.

Disclaimer: I don't own any Tesla stock, neither I have any Tesla products.

It's almost like software has bugs. Even at magical company by the internet's most loved wizard.

Who wants to bet their astro-turfing squads aaa, sorry, PR teams will be all over this downvoting and disliking on every social media platform.

The converse conspiracy theory holds equal validity: that China might have an incentive to generate poor test results for non-Chinese autonomous / electric vehicles. An industry they've publicly stated is a national priority.

Here is a longer video, seems like other cars are able to break before hitting the pedestrian. Also, there is no "still" pedestrian, as some point out.

https://m.weibo.cn/detail/4551307186670198#&video

Still rooting for Tesla to figure this out!

Although I agree with others that Tesla is overselling the current FSD capabilities, it seems that the deep learning community is figuring out 3D video perception (Tesla taking a big part of it, though I'm not sure how many papers they publish)..it improved like crazy last year. Just look at a few videos from two minute papers:

https://www.youtube.com/user/keeroyz

What a time to be pedestrian :D

Maybe this is a software localization?

https://slate.com/news-and-politics/2015/09/why-drivers-in-c...

1. The test may be fake, and Tesla driven manually.

2. Maybe Tesla uses not only cameras but also radar sensors. A blow-up dummy is transparent for radar. Real human is not. So it is OK to drive through balloons.

Car radars aren't sensitive enough to pick up pedestrians anyway. They're useful for detecting metal vehicles, and pretty nothing else. If you mean lidar, dummies are not transparent to those wavelengths.

This is not true. Car radars do pick pedestrians, but some car radars (blind spot for example) filter that out in post processing.

1. the entire point is to test the emergency braking features which are active by default.

2. I would definitely prefer a false positive where tesla brakes for human shaped balloons crossing the street than the potential for plowing through a human because of a bad radar reading.

this video is obviously not unbiased but here you can see a Tesla plow through a dummy that is obviously not made of balloons https://youtu.be/tkAtRw51jGY?t=81

> 1. The test may be fake, and Tesla driven manually.

The second video had in car footage clearly displaying a visual warning and playing a collision warning sound. Because of concerns over phantom braking, these can be overridden by the accelerator pedal.

I wouldn't call it fake, but it is exaggerated.

Ah yes, the random twitter account from China. The best of journalistic sources

Only the best journalistic sources can be trusted even when it's a live video because the entire thing can be completely fake!

I mean who knows, if it's not from a "best" journalistic resource than it could be some random dude in his moms basement spending millions of dollars to create several computer generated videos of a Tesla failing to stop for a dummy. People these days will believe anything.

It doesn't take millions, it just takes a willingness to hold down the accelerator as an override.

Yeah this is why I avoid Teslas on the highway when I'm driving. Their owners are more often than not on their phones not looking at the road while autopilot has their lives (and others) in its hands.

As a former Tesla owner I can tell you we feel the same about humans controlling cars. On the highway between Baltimore and DC there was 6 of us on autopilot, we had FSD and we all kept a safe distance and passed slower cars for about 40 miles.

It made me realize if we all had the same tech in cars in terms of safety and OS, we could be safer than human control and we were.

I wonder if this is a case where camera+radar sensor fusion gets confused. Radar in Tesla's is supposed to be only supplementary sensor but it's there and contributes for decisions.

>Something made of wood or painted plastic, though opaque to a person, is almost as transparent as glass to radar. https://www.tesla.com/blog/upgrading-autopilot-seeing-world-...

This should be concerning to people since there are often wooden barriers around construction sites

With black box deep learning models it's hard to know. DL models learn complex representations but fundamentally what they do is just interpolate between examples. Throw something out of distribution at them and they behave strangely.

For example (speculation):

1. the system knows how to stop when it recognizes stationary radar transparent structures (traffic cones or wooden structures)

2. the system has experience with plastic bags on the road and learned to drive trough them without stopping.

3. the visual system has learned to stop when it sees walking humanoid figure.

The previously unseen rivalry between 1+2+3 leads to delayed decision.

Elon is working on a new "baggo meato sensor" using Neuralink and German shepherd brain tissue to "sniff" if a object is a "real" human being or a fake imitation.

Is there a name for how the FSD feature is structured to never be responsible for accidents? Tesla has so many disclaimers about how you always have to keep your hands on the wheel and be alert, so when accidents do happen they're blamed on the driver. Catch-22 maybe?

Fraud?

Telsa tried to use the "His hands weren't on the wheel, look, he got an alert" argument in the NTSB investigation into an AP fatality. It was later proven that the alert was some 7 minutes prior to the accident.

You should consider whether Tesla considers Autopilot engaged if it was disengaged, say, 10ms before the collision.

It's pretty clear from their long history of weasel-wording that the answer is no, but they also aren't saying.

Unrelated question:

I'm reading this with Firefox on Android. None of the Twitter links work, reporting "Cannot Complete Request

"Additional information about this problem or error is currently unavailable."

Is the problem Twitter or Firefox?

This happens consistently with me. Reload the window where Twitter is appearing and it always loads on the second try.

Try updating Firefox? I've been having this trouble with Nightly Firefox and Twitter for a few days. On desktop, clearing Twitter-related site data fixed it, but on mobile it was only fixed in this morning's build.

Another similar test where the Model 3 did break:

https://youtu.be/cMiZa3HgRVE?t=140

break != brake!

Non-deterministic results aren't exactly a point in favor of Tesla's auto braking system.

A similar article (a bit older but not? Chinese based tests):

https://insideevs.com/news/377427/video-tesla-model-3-failed...

An interesting insight was that pedestrian what most early detected by the Tesla but it just decided to brake only just before hitting them...

Given that Tesla are pre-selling full self driving that will never exist, what kind of financial liability are they exposing themselves to? Do their T&Cs say "Oh btw we might never deliver this. No refunds!"?

For the first video I don't understand how this could have worked to begin with. There's no way my Tesla would have allowed triggering autopilot in a parking lot with so many lines.

Teslas deep learning auto pilot is so smart, that it understands this is just a dummy. So no need to hit the break :-/

This definitely seems like a legitimate test. Especially with the music and everything.

Given that it is similar to the results found by the AAA tests last year [1], I have no problem believing these findings are legitimate and reproducible. Nonetheless, it would be helpful to know the speed at which the vehicle was moving and other such factors.

[1] https://www.aaa.com/AAA/common/aar/files/Research-Report-Ped...

Guys, guys, this is a thing with a lot of cars

Here's BMW, Mitsubishi and others failing at the same thing: https://www.youtube.com/watch?v=7Y8JG7kepwc

The only reason you don't hear about it, is because hating on Tesla gets clicks.

I don't see BMW, Mitsubishi calling their system AutoPilot and their CEOs making all kinds of outlandish claims.

Tesla marketing claims they have a self driving car.

Tesla legal says you couldn't possibly think Tesla has a self driving car.

About as interesting as watching a unit test fail without context.

WATCH: code running in debug scenario fails to run successfully.

WATCH: system designed to stop for vehicles fails to stop for something not a vehicle

tl;dr the "robot" is a mannequin, and "vaporized" is merely being struck.

Given your tldr which presumably summarizes the relevant parts of the story, is it your take that this is a nonstory because the autonomous Tesla simply hit the pedestrian at a high rate of speed rather than "vaporizing" them?

Ah well you see Tesla's system is so sophisticated that it knew that it was a dummy and not a real pedestrian.

If radar is used together with the cameras, then yes, Tesla's system knows this was a dummy.

...and hit it anyway?

Reply on news.ycombinator.com