Self-Driving Truck Encounters the Unexpected, Then Reaches a Milestone | Hacker News Reader

Self-Driving Truck Encounters the Unexpected, Then Reaches a Milestone

blog.caranddriver.com

38 pointsedward60 comments

Honestly, I do not consider stopping squarely in the middle of the highway and blocking one of the lanes as the "correct" way to handle this situation. I'd rather expect that the truck at least knows how to stop in a safe manner, i.e. by moving to the shoulder first. A large number of accidents (probably even the majority) here in Europe are caused by drivers crashing into stopped vehicles on highways, so having AI-controlled cars stop in the middle of the road is clearly not acceptable.

Cannot agree more. Another possible improvement should be stopping in the next gas station or resting area, so the truck always keeps enough internal data to move up to the nearest safe place to stop. Sort of a buffer that allows the truck to be minimally operational, maybe also a significant decrease of speed.

Agreed, but I would imagine a meaningful population of people would do the same thing in that situation

Maybe. But the promise of autonomous cars is that they can be safer than human drivers.

“That situation” was loss of contact with a control center. What is the equivalent for people?

fainting. I do believe braking and stopping is better than swerving and crashing...

no, there isn't a person constantly driving on the other end of the connection

Perhaps dense fog, or whiteout blizzard conditions? I’m making the assumption that the server parses the data from the truck to determine what it is seeing.

No, those are sensor failures, not external communications failures.

The human equivalent would be being unable to call someone for directions.

The article says it's a person making decisions with acceptable delays of 15 seconds.

Sure, some small % of the populations will probably do that. Some small % of the population will drive drunk.

Neither are acceptable ways to handle the vast majority of situations one encounters.

The situation in the article. is roughly the same as being lost (or going somewhere in a way GPS won't work) and taking directions over the phone. If the line goes dead nobody (who's behavior we want self driving vehicles to emulate) just stops in the middle of the road.

You are absolutely correct. Here's what can happen if you stop in the lane:

http://www.bbc.co.uk/news/uk-england-beds-bucks-herts-433085...

It's a little terrifying that they think this is the best way to deal with the situation. And that's without addressing the elephant in the room - a loss of comms was so unimaginable that they had no plan to deal with it?

Worse, if they didn’t account for this rather obvious situation, what is going to happen in the dozens - hundreds - of other corner cases?

In my view, stopping in the lane is the safest thing to do if the truck suddenly doesn't trust its own navigation system for any reason. Hopefully those reasons are, and become increasingly, rare. What I gathered from the article is that the truck slowed down gradually and stopped. For better or worse, this is something that everybody else on the road has been trained to handle, even if it could be a monumental inconvenience.

The article keeps iterating how they’ve tested for every conceivable condition they can think of, and yet the narrative says that they were stunned to experience a communications loss which they never tested/conceived of. It’s not reassuring when someone claims 100% test coverage and 100% pass, when they only have a limited number or tests in their script. A comms loss was unthinkable?!? What is this, amatuer hour?

They say that the truck/system is autonomous?

Because, yes, if loss of connectivity is a major fault and put it into a safe mode, then no, is not autonomous driving.

Agree. It is "automated" driving, not "autonomous" driving.

My thought exactly, I'm sure Boeing doesn't consider a loss of communication unexpected when they develop Auto-pilot systems.

100%. I hope they spun it as "unthinkable" for PR reasons, which would be slightly less bad.

The article is very thick in drama and linkbait, it really could've been written much shorter - "autonomous truck stops because the server went down, welp guess they didn't test for that". Except that they probably did because the truck stopped safely.

It went into "safe mode", which sounds like a catch-all for every conceivable and inconceivable error they don't explicitly handle in any other way.

This is a Systems engineering miss. My guess (would be interesting confirm) that this is a "software first" culture, which can result in these sorts of things. My brief linkedin search of the company showed UI, computer, and data engineers... but I might of missed the Systems.

Came here to post the same thing. For any developer who has worked on network software longer than 10 days, the complete loss of signal to the "server" is _not_ "an edge case"!

On one hand, it's great that the Semi truck was able to slow and stop without hurting anyone, after losing connection with the control center.

On the other, how does this bode for an autonomous fleet of trucks that all suddenly stop somewhere along a highway? The traffic disruption could be nightmarish. And what if there was no lead vehicle to go fetch said truck in a timely manner?

Also i feel like there is a lesson to learn here about how it's not just hacked servers and software that could produce these events, but a hacked or disrupted (lightening striking a utility pole) power supply could also have immediate real-world impact on these transport systems. Particularly if you dont have backup control that is distributed and redundant (or onboard the vehicle itself).

That all rather depends on why the connection was required, if it's a fundamental part of the planned way it'll work in the future, sure. If it's a current safety requirement while testing, it's less of an issue.

This is the tradeoff we get with technology. It gets increasingly better with every iteration but the consequences of failure gets worse and worse. What happens with humanity if a big solarstorm comes again?

We are incredibly eager to introduce technology dependencies it seems. The idea of robustness via simplicity seems to be a mantra for individual bits of software, but we are happy to stack it all together in a teetering jenga tower of dependencies. We rave about how great it is to have a good ecosystem in a platform or language, but the word ecosystem is synonymous with complex dependency trees (ha!). Nature is definitely a more robust architect than us though.

As for what I would do when it all falls apart, I would barter for a sailing boat and some paper books on how to sail, and live the life of an ocean vagabond until I inevitably perish on some island or at sea.

So, it lost its umbilical to the great brain in the sky and came to gentle halt on the highway.

Color me unimpressed.

Sorry for my snark, but, donkeys do better. If only they could travel at 70mph :/

We are making incremental progress towards this autonomous vehicle goal, but hey, let's get some perspective and not let our techno-fetishism get us overexcited. Trains and stuff.

So you're advocating taking a donkeys' brain and hook it up to a truck and let it drive?

I hope you're being facetious. I am not particularly advocating anything, other than to point out, you know, we could have invested in freight trains for the past century and this would be less of a problem.

I wonder if modern lifesupport systems could keep such a constellation running for long enough to deliver goods...

Maybe an insect brain is enough, the ones that fly are very fast in reactions.

My windscreen would disagree with you ;)

You hit the insects with your car, also you need to consider distances relative to the insect size.

I wonder if a real brain ( animal, insect, etc) + all senses, could be trained to "stay in the line" and use a interface read out the reactions ( like, there is an obstacle, or change direction ) and translate this into commands given to a vehicle.

Humas do this when they learn to drive, but why only humans shall learn this? Is this so complex?

They got a rat brain to fly an F-22, so sure, why not?

https://www.newscientist.com/article/dn6573-brain-cells-in-a...

Better than old University professors, as in Philip K. Dick's Mr. Spaceship!

I mean, I'd prefer it if it pulled over at any potentially unforeseen problem with its hazards on, rather than going "naw let's do this" and continuing. I don't see why that's worthy of condemnation.

Is it self driving if it depends on "the cloud" , next time they will get stuck in a tunnel or if the internet goes down.

I think the OP point is that is nothing impressing or that makes the reader feel more safe with this kind of trucks around, having a safe mode is something normal that they should have is nothing special.

If it was a human driver stopping in the middle of the road for no good reason, the driver would likely be cited by the police and fired from the job.

Forgot where I read it: self driving cars should not compete with human drivers. They should compete with the horse. A horse will generally avoid obstacles, won’t crash into things and cause damage, and if you get blackout drunk, will walk you home. They are the original self driving cars.

> the truck inexplicably slowed and stopped in the middle of the road.

This scenario caused a fatal crash in England which is now finishing up in court. 8 people dead.

http://www.bbc.co.uk/news/uk-england-beds-bucks-herts-432271...

I wonder if these vehicles are performing advances regarding other vehicles they encounter. If that is the case, this particular truck could stop in the middle of the left lane, can you believe that?

Oh Fuck No.

The "safest decision of all" is most definitely not to stop "in its lane of travel". That's insanely dangerous in any real-world scenario.

At least have the vehicle pull over onto the shoulder. What makes them so "special" that they think they should just stop in the middle of a highway? If that's the state-of-the-art of self-driving, then those vehicles simply aren't yet ready to be on the road.

Some time ago there asked people how should self driving car react in various situations if collision is unavoidable (e.g. should they run over old man or young girl) but the actual algorithms are much much simpler: stay in your lane, if you see obstacle break.

In the context of this story, “break” is an amusing homonym for the word you probably meant (brake).

It should actually not be "insanely" dangerous to stop at any point, if you have sane drivers and laws. You should always be able to stop before the visible part of the road ahead of you, or before the vehicle in front of you, right?

But yeah stopping on the shoulder of course is a no-brainer. How to do that, though, is maybe not so simple. In winter judging where the road ends and a ditch begins can be very hard if there is no rail.

> You should always be able to stop before the visible part of the road ahead of you, or before the vehicle in front of you, right?

...unless you consider that "how about winter" part you used to explain that pulling to the road's side is not that simple.

> You should always be able to stop before the visible part of the road ahead of you, or before the vehicle in front of you, right?

You're following another vehicle at safe following distance. Which is much shorter than distance you can stop at if that car stopped in 0s. Car suddenly changes lanes, revealing obstacle in your line. If you can't change lanes as quickly for some reason, you've a massive problem...

For exactly the reason you list, what you describe is not a "safe following distance."

You can argue the point until blue in the face but you're not going to improve anyones safety by chasing this argument. Humans are terrible at managing risk. Humans will always make mistakes. Telling them to 'not follow closely' is futile.

If you don't design systems to take account of this then your safety practices will be ineffective.

If autonomous systems are to be on the road with human users then those autonomous should be designed in a way to minimise the mistakes humans make.

They should be designed that way because they can be designed that way.

I agree completely, and I never advocated for any change. I do, however, like to point out cognitive dissonance when I notice it on HN.

It's not cognitive dissonance at all. You're talking about safe speed based on line-of-sight. Meanwhile safe following distance is completely different matter. One of the reasons why traffic works is that we trust other drivers and rely on them.

For a car traveling at 70 mph, piloted by a human, it's estimated to take about 100 feet for them to react in an emergency, and another 250 feet to stop[1] (although theoretically high performance cars can brake faster, of course).

So while you can imagine everyone leaving over 20 car lengths between them and the next vehicle on the highway, it would be a radical and probably infeasible change in driving behavior. Not to mention how much slower people would have to go around curves if they didn't assume a lack of obstruction.

[1] http://www.government-fleet.com/content/driver-care-know-you...

It is safe following distance if you can stop if the car in front of you stops. As in, you have to have time to react to the other car braking and then you have time to slow down while that car is braking. Not if a standing obstacle comes from behind that car in no time. Especially if that obstacle is not visible through the car in front of you.

After all, traffic rules clearly say "no stopping on highway" specifically for that. Even shoulder is only for emergency and you have to GTFO ASAP from there.

> But yeah stopping on the shoulder of course is a no-brainer. How to do that, though, is maybe not so simple. In winter judging where the road ends and a ditch begins can be very hard if there is no rail.

If the vehicle can't handle that then it can't handle driving without a human on board. Being able to safely come to a stop somewhere that is not in the middle of a traffic lane is table stakes for any unmanned vehicle.

This is eerily like the common sci-fi movie plot of killing the “brain” to eradicate the entire fleet of bad guys. (Independence Day, Star Wars Ep 1, etc)

Imagine what this would be like at scale. Besides the danger of having thousands of trucks suddenly stop on the highway, now whatever those trucks were actually shipping is also stuck.

Should autonomous cars have multiple self driving systems, so if one service is DDOS'd, they can failover to another?

But autonomous vehicles are not autonomous (as in "without dependencies on a complex chain of tech with human operators in it"), and all I can think is “hackable."

Odd how thoroughly the original meaning of "milestone" has been lost.

Reply on news.ycombinator.com