Hacker News Reader

Comment by melling | Hacker News Reader

“the truck drove mostly autonomously.”

That last 10% is going to be a lot of work.

https://en.m.wikipedia.org/wiki/Ninety-ninety_rule

I'll take a 90% efficiency gain any day. If they can drive I-70 from SF to NYC with no human, or the 10 from Florida to California, that seems like a win, even if the person has to drive in the city. Warehousing will just get shifted around to work with where the trucks can safely drive.

I'll take a 90% efficiency gain any day.

If you need a safety driver 100% of the time, the net gain in efficiency is 0%.

Edit: There's no comment in the article about there being period with no safety driver.

If it's one safety driver monitoring 4 vehicles it's still a win.

But that isn't what the article talked about either (A European company did do a run like that a while though).

It did talk about "level 4 autonomy" based on a single drive with no disengagements (they mentioned they'd made a lot of drives but didn't mention their disengagement rate overall), doesn't sound like "achieving level 4" autonomy either-either.

How is one driver going to take over for 4 vehicles? Each truck needs a safety driver.

90% less human mistakes.

I'm sure someone's working on telepresence which kicks in when the AI's getting confused.

Earn credits for Stadia - solve these realtime truck simulator captcha!

I would imagine that the current google captcha system has so much traffic, you could use it to power real time driving with static pictures. Set 20 people to tell you if there is a stop sign. 50 more to decide if that object in the road is dangerous or not. All you need is a decent number of responses coming back that agree.

This is halarious and terrifying. I can envisage groups of trolls who get together to captcha-troll lorries into buldings or worse for lulz and the terribly written legislation shoved in place to mitigate it before nation-states get in on the fun.

>> If you need a safety driver 100% of the time, the net gain in efficiency is 0%.

If this trial period helps us improve the technology and eventually remove the safety driver after 1, 2, or 10yrs, that is a huge gain. What is the rush? -- every technology takes time to perfect. Lets judge the technology once it has had time to mature.

If the truck still requires a human for some tasks (intra-city, fuelling, etc.) but can be rated to legally drive itself for most of the trip, then the human could sleep / eat / etc. during those times they're not required. This would presumably let them avoid minimum stationary rest times etc. since they could get a full rest of sleep while the truck drives itself (on sufficiently long journeys).

For owner-operators who live out of their vehicle anyway and are more concerned about getting work than sleeping in a hotel somewhere this could dramatically increase the volume of work they can take on (at least until fully autonomous trucks become widely available and humans are completely deprecated as truck drivers).

From that perspective it would be more efficient, not 90% but there would be significant gains.

Well, if you had trucks that were safe enough to sleep in while they auto-drove, the companies would certainly want to go one better and have them drive empty. They could create dedicated fueling and loading areas if need-be.

But the trucks being that safe is the big question. The article describes a single trip without disengagement. Much more that is clearly needed.

"most of the trip"

So if I get woken up every 10 miles, am I really meeting the rest requirements?

I was thinking more like "the truck can legally drive itself until the next fuel stop in about 14 hours time, so I'm going to get some sleep".

No.

we still won't allow trains to operate without onboard supervision and they should much easier to automate

It doesnt help that the FRA is pretty awful and stuck in the 1880s.

Handling that 10% is going to be tricky, though. Can an unpiloted truck be trusted to know its limitations perfectly, shutdown and park when it hits borderline road conditions?

Edit: Okay, remote driver access might solve this.

Judging by how frequently I occasionally lose internet connection for a mile or two in my major metro area for even basic tasks, like Siri dictation or streaming audio, I can't imagine how remote takeover could possibly be the answer to the "10%" situation unless we're prepared for some percentage of the time we are faced with a truck that simply gives up, puts on four-way flashers and comes to a stop wherever it may be (which could include network dead zones)... and then it's a matter for the police and/or recovery workers to either get the AI back in the saddle or move the truck.

> Judging by how frequently I occasionally lose internet connection for a mile or two in my major metro area for even basic tasks, like Siri dictation or streaming audio, I can't imagine how remote takeover could possibly be the answer to the "10%" situation

I mean, it's possible that demand for self-driving trucks means reliable wireless connections get implemented everywhere they can drive.

There is already fantastic LTE along all the long-haul truck routes, because human truck drivers like to be connected to the rest of society while they are alone on the road for dozens of hours.

Also, Starlink is coming.

Interesting points.

That would be quite the feat given that I70 goes to neither of those states.

I80 goes SF to NJ, but no single interstate will take you from SF to NYC proper.

I'm pretty sure the control city on 80 east is NYC and not Newark.

https://en.wikipedia.org/wiki/Interstate_80#New_Jersey

"I-80 does not enter New York City. [...] I-80's designated end (as per signage and New Jersey Department of Transportation documents) is four miles (6.4 km) short of New York City in Teaneck, before the Degraw Avenue overpass."

The problem is that you can't plan the remaining 10%. Perhaps it's every 10 miles you have to have the safety driver take over. That makes it effectively useless.

The devil is in the details... Which are totally absent from the article.

So what if it drove itself down the freeway? This is literally the easiest thing that "self-driving" vehicles manage to do.

Did it fuel itself? Did it take exits and drive itself to the gas station? Where did the driver take over when it got near the delivery docks? Did the truck also get clearance and dock itself? Did the truck even pull itself out of the loading dock when it first embarked on it's trip? Those are the hard things to solve, and conveniently are omitted.

I want to hear more about how it will work when guys who are braking-bad can drive up alongside the truck and determine that there's no human driver. One car in front, one car on the side ...

This would still be a big win for long haul trucking, most of time is spent staying in lane on the interstate. The drivers are required to take breaks to stay alert which adds time to the delivery unless alternating drivers.

IMO this is where self driving should focus, not city driving. Truck drivers or anyone gets the vehicle on the interstate and you can sleep or whatever to the next exit for fuel or your destination then take over again.

> The truck... had to take scheduled breaks but drove mostly autonomously. There were zero “disengagements,” or times the self-driving system had to be suspended because of a problem

These seems to imply short bursts of voluntarily shutting off auto-drive; I'd be curious to understand the rationale for when the two human copilots did so.

it's worse than that, of course. You'll find, like a certain Greek philosopher fond of asking annoying questions, that the first 90% of the last 10% will be OK, and that last 10% will be the other 90% of the ... etc. etc. etc.

That’s still a hell of a savings.

In 1995 having a domestic phone call, placed on a wired line, drop due to line issues was effectively inconceivable -- it simply would never happen. In 2015 we were in a time when it happened all the time, people simply accepted that voip was often unreliable.

In 2019 the idea of a vehicle around you simply stopping because its driver doesn't know how to handle a situation is extremely rare.

I suspect that by 2039 it will be far more commonplace, shrugged off as just a fact of life that electronic drivers fail on a regular basis and need to pull over and wait either for intervention or conditions to change.

True ... I kind of forgot how reliable analogue tone switching were. Someone cut of the wire once in the road. It's like I see more and more cars stuck at the roadside. Atleast in the 90s the drove until they started to smoke.

Is that actually true? My parents had phones around then and they've always had this habit where they say "Hello! HELLO!" At the shortest momentary pause in the conversation.

No, we had calls drop often enough when I was a kid. But we didn't live in a big city, and it wasn't surprising to us when it happened.

It happened about as often as it happens to me with a cellphone when neither party is moving between towers and has decent coverage.

We expect cell drops because of a technology constraint, even in big cities. Likewise, I think the parent poster is right that we'll expect it more once it starts happening more.

Huh? The comment you're replying to says the first 90% of the work is easy and the last 10% is difficult. What you're saying is very much not "worse than that"; you're saying the first 90% is easy, then it's a small jump to the first 99%, and then just a bit more to the first 99.9%...

That's a huge improvement over the last 10% being difficult!

Reply on news.ycombinator.com