Hacker News Reader

Comment by brainless | Hacker News Reader

"You are still responsible for your car and must monitor it and its surroundings at times within your line of sight because it may not detect all obstacles. Be especially careful around quick moving people, bicycles, and cars."

Is this a joke? I have huge respect for Tesla and SpaceX, but how do they expect to release a feature that will self-drive a car (even tiny distance) and someone will oversee a moving car?

The software engineer in me just died.

View on HN

>Is this a joke? I have huge respect for Tesla and SpaceX, but how do they expect to release a feature that will self-drive a car (even tiny distance) and someone will oversee a moving car?

Think of it as a marketing/business idea (let's sell this "advanced technology to early adopter suckers and we can try and spin any accidents as operator fault") rather than an engineering idea...

Except with the added bonus that some people might get injured or killed. When will software engineers take this shit seriously?

While at the same time "brings your car to you while you are dealing with a fussy child". It's like marketing wrote one part, legal the other, and neither read what the other wrote...

Disclaimer: this product is completely useless.

I think Tesla may be setting themselves up for trouble here. These sort of “using the product as clearly intended voids the warranty” disclaimers often don’t hold up.

I agree. Why even bother using a feature that cant babysit (manage) itself. Yikes. I rather it not be enabled at all if its going to risk my vehicle. Thankfully I am letting the masses QA those cars. I will stick to my dumb car in the meantime.

If you walk through parking lots, you unfortunately don't have a choice but to be part of the QA process.

This is very true. Sounds like a class action waiting to happen.

It's not unreasonable to assume someone will have a line of sight for where a vehicle is being summoned from, just like when in the vehicle itself and you drive forward you're paying attention to its surroundings as well.

I have seen some videos released with summons that just a quick horn honk would have likely prevented another driver backing up their vehicle from bumping into, with enough force to do some front bumper damage, a Tesla being summoned in a parking lot; it moves slowly too, and that could also be a factor in surrounding drivers not gauging or having the same kinds of reactions.

I can have line of sight of my vehicle, but in a parking lot, there may be a small child, obscured by other vehicles, in the path of motion of my car.

What do you think is going to happen, when a Tesla runs one over?

Hint: They are going to blame the driver for using this feature in a parking lot. Their fanboys will defend the hell out of that statement. They will be oddly silent on the subject of 'If you're not going to use this feature in a parking lot, where the hell are you supposed to use it?'

A 'self-driving' car that is incapable of collision avoidance is not yet ready for the market.

Ever walk up to your car in the parking lot then realize... that's not your car? Now identify your car from across the parking lot. In California, where every other car is a ticky tacky Tesla (and they all look just the same). There are already anecdotes of people wondering why their car isn't summoning only to realize ... that's not their car.

Fair point - so certain scenarios likely would benefit from visual and perhaps audio signals, lights flashing - certain lighter horn noise. Perhaps even listening to audio for someone nearby to be able to say "stop" and the vehicle would stop - or respond similarly to other horn noises.

New electric cars have a low speed noise maker in the US (required by law after sometime this year), but most Teslas on the road now were built before this was a requirement.

I think that's one of the stupidest additives of constant noise pollution - noise pollution being one reason people zone out and stop paying attention to their environment, rather than people being aware that they're about to step onto a road.

As someone who grew up before smartphones, with plenty of noise pollution, I am sure it's not the noise. People cross the street stairing at their hands.

Not everyone can see.

And those people are going to be much less likely to randomly, unexpectedly walk out in front of traffic where it's not a designated crosswalk. I bet their hearing is refined, heightened, to more easily notice quieter electric vehicles as well.

I think the requirement isn’t in effect until 2020:

https://www.theverge.com/platform/amp/2019/7/1/20676854/elec...

> In California, where every other car is a ticky tacky Tesla

I live in California, and if I see five Teslas in a day it's an unusual cluster.

Now, Silicon Valley probably has a greater proportion of Teslas, but Silicon Valley isn't typical of the state.

Really? Even just during my 10 minute walk to work in Minneapolis, MN I'll probably see five every day.

I’m in Orange County and I usually see 8-12 on a short drive. Easily 30-40 across a normal day with a 15 minute commute.

Live in SV and my commute i see average of 30 M3s, about 10 S and 5 X mostly on 101. Also i see about 5 delivery truck packing about 5 cars each every day. Thats a lot of cars.

Why do people say that highly populous areas are “not typical”?

Median household income in San Francisco-Oakland-Fremont metro area is over $100K, for the state as a whole it's about $70K.

That's one reason the former isn't typical of the latter.

>It's not unreasonable to assume someone will have a line of sight for where a vehicle is being summoned from

And do what? Magically jump in the car and take over the breaks when a kid suddenly appears on a bicycle going towards the car?

Or are they only supposed to summon the car in the Mojave, where there's fully open line of sight and few passers-by?

I think the summon feature requires user to keep pressing a button on the phone to work. Lifting finger from the button will stop the car I believe.

I’m saying it by watching the videos. Someone knowledgeable do correct if I’m wrong.

That doesn't stop Tesla from driving into live traffic and stopping there. Look at the videos.

So instead of a foot on a pedal connected mostly directly to the brakes, it is a finger on the button of a phone app over the internet? That is madness, imagine if your car's brake pedal had to go over the public internet and instead of a pedal is was a button in a probably poorly coded phone app.

> So instead of a foot on a pedal connected mostly directly to the brakes, it is a finger on the button of a phone app over the internet? That is madness, imagine if your car's brake pedal had to go over the public internet and instead of a pedal is was a button in a probably poorly coded phone app.

it's a deadman's switch not a brake pedal. if you lose connection it stops the car. if you lift your finger it stops the car. if the car senses an object it should stop the car (but might not because it's not 100% yet)

What’s the latency tolerance for this? I have to imagine that, given mobile internet conditions, it’s in the range of seconds. It also raises the question - is it really a dead-man’s switch, or more like a heartbeat being sent every N milliseconds? How many heartbeat signals can be lost?

It’s not as simple as a deadman’s switch; it can’t be if it’s using a mobile phone and the internet.

This is a false problem, the engineer have coded the button the other way: Only moves if you receive a signal from the app. If you lose this information in less than 50ms, just stop the vehicle.

It's not a solvable problem in this context. Humans are not computers, and cars have no self. The car would melt a hole in the ground and not even approach par with a cat. It's understandable that people are making this mistake in the context of lifetimes of sci-fi anthropomorphizing machines.

You are correct.

No, they can stop the car pretty much instantly. No magic required.

And for the case you mention, a kid on a bicycle, the car will stop itself if needed for safety.

But magic _is_ required. A reasonable field of view to assess the hazard or approaching kid on the bike might, quite easily, be available from the car, but not 20m away looking toward the car.

Case in point. That car in the video that turns into the car park looks like comes from behind the guy with the phone. Someone in the car, or the self driving sw, should have seen it far earlier. Maybe even noticed an indicator, or deceleration.

And that scenario of a kid on a bicycle not being seen until they're almost in front of you exists while a driver in the vehicle too: collision detection will certainly be able to have a faster reaction than a human driver as well.

Fortunately the cars do have cameras facing in every direction. Clearly not sufficient to stop every fender bender but you could say the same about human drivers.

The car should stop itself for safety, but may not have the awareness to do this successfully. Bicyclists are explicitly listed in the warning about paying attention to the surroundings of your car while it is being summoned.

Paying attention to surroundings is good.

>And for the case you mention, a kid on a bicycle, the car will stop itself if needed for safety

That's what it's reported as NOT doing, tho (and also what Tesla asks you to keep an eye on yourself while summoning, in their small print)

Is there video of such an incident - did you see a video of such an incident? Not saying it's not possible, however we have to keep in mind what's "reported" as clickbait "journalism" vs. reality.

It wasn’t even reported... he/she’s just making stuff up.

First of all, that's an ad-hominem (and rude).

The title of TFA is about how "People Are Already Reporting Collisions With Tesla’s Driverless Smart Summon Feature".

So clearly it's not the case that "the car will stop itself if needed for safety"...

Your point is that if it was for a kid on a bicycle it would, but for the reported situations (not being kids on bicycles) it doesn't?

Because the more obvious conclusion is that it doesn't work properly period, and that can bite people, where the case involves an incoming bicycle or not. Especially since we have reported cases of Teslas hitting bicyclists in the non-summon mode:

https://www.youtube.com/watch?v=I3m1U2KiGgI

> That's what it's reported as NOT doing, tho

False. Nowhere is that reported.

Yes Tesla advises you to be careful.

If you think harder about not, no one really turns on autopilot except on main roads or highways, hence almost no training data. Parking lots is like a big rectangular area with stuff moving all over. So with more training data, they should have it down, but these incidents gonna lower their rates of getting training data.

Is a chicken and egg problem.

And if the pharma companies didn’t have to abide by ethical guidelines for clinical trials they could collect more data. There are good reasons to forbid that.

I thought Tesla collected training data everywhere all the time?

And they shadow test while hooman is driving.

of course its unreasonable, but thats a liability clause added by a lawyer.

How is it unreasonable if the things are already having collisions? This looks like a half-baked feature not even close to ready to be pushed out. Tesla is becoming a meme at this point and it has to be coming from the erratic man at the top. An engineer worth anything wouldn’t launch something like this.

Sorry if I’m fired up about this but he just reminds me of a bad boss I had that only cared about hype and how things looked and expected the laws of mathematics and physics to bow to his delusions.

It’s possible many of these collisions are caused by other drivers.

I mean even parked cars get hit by drivers every day.

There is a bit of news hysteria effect happening here.

There’s literally only two examples in the post and one is an example of the car stopping before a collision. The other example is another car backing into the Tesla.

In the first video from the article, the Tesla would be found to be at fault in most jurisdictions as the driver backing out was well into the backing out maneuver before the Tesla even rounded the corner. The Tesla without question did not have the right away at that point.

In most jurisdictions, the vehicle traveling in reverse is always the give-way vehicle. Practically, I expect that the insurance companies would agree to each pay their own driver’s costs and count it as an at-fault accident on both sides.

A self-driving car's fender bender is more newsworthy than a fatal accident with a human driver.

As you should know as a software developer, development of major new capabilities is an incremental process, and for something like self driving cars it’s expected that there will be a transitional period before self driving. Summon in parking lots is a good transitional step.

Remember Tesla is not just training cars here; humans also need to learn to deal and not freak out. There will be fender benders along the way. Most of them probably caused by other drivers, but bugs are also a possibility.

>As you should know as a software developer, development of major new capabilities is an incremental process

Hopefully software developers with such an idea are not allowed in medical devices, aviation, missile guidance, space industry, industry robotics, and automobile...

Developments in all the areas you mention start with baby steps and incrementally progress toward greater and greater capabilities. This should not be news to anyone here.

Development yes. Releasing them as products (which is the case we're discussing) when they have people-killing faults, no...

Plenty of people die in cars every day. Cars are released products, right? You are making no sense here. And the collisions reported with smart summon are fender benders not “killings”. No need to hype up the drama level.

>Plenty of people die in cars every day. Cars are released products, right?

Yes, but current cars killings are due to operators (drivers) errors. Not in a self-driving parking-summoner killing people.

Companies don't (or aren't supposed to) release production cars with known people-killing faults. And when some are nonetheless released (which happens sometimes, e.g. a defect found in the break system) companies get fined and/or people go to jail for those. It's not all A-OK because "people die in cars every day" anyway...

>You are making no sense here.

Oh, the irony.

Sure, but usually we don't let Joe Blogs control it or even get in it or near it until there's a level of reliability and safety in its behaviour.

I think you’ll find that the behavior that is unreliable here is that of other drivers, not the technology. So what you’re saying doesn’t apply in this situation.

Even if it did apply, I believe the technology is safe and reliable, because from what I have seen Tesla does extremely extensive testing. So this meets your filter as ready to be released to the public.

Of course it applies!

Human drivers are reliable in the presence of other, unreliable drivers. We deal with this kind of uncertainty every time we get behind the wheel.

It's not a stretch to say that we should expect at least that much from automated driving.

To that point, I believe this technology was not released willy nilly. The poster seemed to be assuming it was. I agree that a good level of safety and reliability is important.

Nothing has ever been or will ever be perfect revision 1. That doesn’t mean it shouldn’t happen. The real world will always be where flaws get light shined on them.

To your point, no one is still using Model T’s, flying Wright Flyers, or still living with a Jarvik 7 heart implant (note the 7 implying incremental progress). Further, the entire history of space exploration is built on learning from failures and explosions.

If we replaced 'Tesla' with 'Boeing' and '737 Max', would you still feel the same?

The "humans are the bugs" excuse is pure denial on the pre-programmed proponent side.

Nobody called humans the bugs.

But sure, humans sometimes do make mistakes.

And when a Tesla is involved on either side, news hysteria effect seems to kick in.

Reply on news.ycombinator.com