An ex-trucker has some questions about the Tesla Semi | Hacker News Reader

An ex-trucker has some questions about the Tesla Semi

autoblog.com

71 pointsranit82 comments

A bit of background: I've done thousands of hours of driving in rolloff trucks. They're semi's that load debris containers onto rails mounted over the chassis, haul the load to the transfer station/landfill, and deliver containers to job sites. Now I write software for the industry that I used to labor in.

I completely sympathize with the concerns in the article, but I have faith that Tesla will be able to overcome the challenges and some of them are perhaps not looking at the trucking industry "of the future" here's a good example:

The concern about a roll down window isn't that big of a deal. Doing rolloff meant crossing a scale multiple times per day where you'd exchange paperwork with the scale house attendant. I just completed a project that will keep drivers in the cab while at a landfill/transfer station if they have a tablet and the site is connected with our scale software. All of the interaction is done on a tablet while the truck is on the scale, the driver never leaves their seat or rolls down a window. If we're doing stuff like this for garbage, surely other trucking is going to be be digitizing these types of interactions and there are some nice big screens in the truck to make it easy.

The biggest concern that I share is the passing issue with a center seat, but that should be easily solved with a camera that shows a forward view when a turn signal is engaged. Two big screens mean you'd get a forward and all the blind spot views necessary to make a safe pass.

Side note: If Tesla wants to make a version for hauling trash, they could have an enormous impact on local emissions.

What's stopping a tractor/trailer from weighing itself, if a large manufacture r was to pay a very large bond to ensure that spot checks were valid?

I was suprised to see there are 13,000+ freight brokers in operation. Where's the Uber?

There's a company that builds electric drivetrain conversions for garbage trucks: https://electrek.co/2017/06/14/all-electric-garbage-truck-ca...

Re: central seating position. Try driving a combine harvester or an agricultural tractor pulling a long train of equipment. They can be heavier, longer, and/or wider than a semi. They use a high central seating position for a reason.

Re: mirrors. They're not there for aerodynamic reasons. The semi has a lot of cameras, so you'll have to use the screens. Sometimes a mirror or sticking your head out the window might be preferable, but it's a reasonable substitute.

Re: night mode. It'd be nice if they used OLED screens with a dark theme so you can really crank down the brightness and black is completely black. Still time to change that if it isn't, so hopefully somebody from Tesla is reading this and/or advocating for it.

I was thinking the same thing about the mirrors. Physical mirrors have their own limitations (poor in low light, poor when there is a badly placed bright light due to glare, etc), its just that we are used to dealing with them. Tesla's implementation may turn out to be better than that. Of course, it is also possible that Tesla's implementation may turn out to be worse, and his article was phrased as a list of questions that he would like to resolve in a test drive.

A camera is preferable in some situations, mirrors and sticking your head out the window in others. Why can't we have both and use the best as the situation demands?

> Try driving a combine harvester or an agricultural tractor pulling a long train of equipment.

Those are not typically driving on roads though. The seated position next to the center line means you don't have to pull as far into the other lane to see oncoming traffic.

Semis pass too.

WRT mirrors: no matter the high tech solution, having a manual backup is a prudent precaution. A mirror is harder to take out with a single splash of mud. It will also ease the transition between driving normal rigs to driving the Tesla rigs. Drivers working within their comfort zone is important when they're running for 10 hours at a time.

[EDIT] Wrote this before reading the article; frankly, the author does a much better job explaining why so many mirrors are a bonus (it's not just behind the truck that matters), and covers the central seating explanation much better. Short of being a trucker, I'm not sure how we can contradict his comments.

> A mirror is harder to take out with a single splash of mud.

Never mind mud, what about the ice? I drove trucks for a while and the entire front of my truck was routinely coated with snow, slush, and ice. If it was heavy enough I'd have to pull over and tap at it with a 'tire thumper' to knock huge chunks off periodically. Keeping the windshield clear was a constant battle. I wonder what kind of camera or sensor could be used that wouldn't ice up. Traditional mirrors have heating and most are enclosed in a front-facing housing. They get really hard to see out of too, because the salt and sand spray kicked up by the vehicle in front of you sticks to the heated glass of the mirror and collects there. Visibility in winter driving is a difficult problem but perhaps that can be improved.

I'd expect cameras to be easier to keep clear in extreme circumstances, actually. The key is the "clear view screen" [1], a circular window on bearings that spins rapidly to throw off obstructions. You can also heat them to deal with icing. I don't think that I've ever seen a mirror outfitted with one, but cameras are small enough to make them trivially fittable, and the units themselves I'd expect to be surprisingly cheap - armored ten-inch units for CNC enclosures are ~USD1300 even at low volume [2]. I could easily see a configuration that has four or five important cameras covered by spinners and placed so ice can't bridge the panel.

[1] https://en.wikipedia.org/wiki/Clear_view_screen [2] http://visiport.com/pdf/VPO-DEU_Unit-Pricing_PL1601.pdf

Still he has a good point about not being able to open the window to reach out to hand papers / etc...

I don't know how it works with "normal" trucks, but won't this mean having to get out of the vehicle to pay tolls, hand license to police, etc?

I don't know how the rest of the country (and world) is but anecdotally in New England the tolls are almost entirely electronic.

Some (most?) exits don't even have manual tolls at all, you have to go to the website and pay within 7 days if you go through one without the transponder.

Getting pulled over by an officer should (hopefully) be infrequent enough where getting out of the vehicle is not a big deal. Although, I know police officers get very nervous if you motion to get out of your vehicle (understandably) so some communication is in order.

Getting pulled over by an officer should (hopefully) be infrequent

I guess it depends a lot on where your route goes. If it's a harbor, airport or any sort of restricted area, you'll be 'pulled over' every time you enter or leave.

Very good point. I hadn't thought of that.

Also trucks have to pull over at weigh stations all the time, and when there are no weigh stations they'll often be pulled over by troopers to check their weight.

Getting pulled over by a trooper is rare, even if you are speeding (unless it's flagrant). It does happen but they tend to focus on cars and I think this is because the ticketing and inspection is easier for them. I've had to get out of the truck and help them find the trailer registration documents in the nose box for example, and I could tell they were a little embarrassed by this.

Weigh stations are simple, you just idle slowly over the scale plate(s) and there is an illuminated sign board ahead telling you to go, stop, or pull around for inspection. New Mexico is the only state I encountered where you actually rolled up to a trooper in a window up at your eye level and had to interact with them, but this was just to hold up a copy of your permit and they waved you through. You didn't have to speak.

Are you comparing a semi truck driven at 70 mph in between two lanes of traffic to a piece of farm equipment that maxes out at 15mph and is typically driven without another vehicle in sight?

His objections are about low speed manoeuvring: backing up, navigating a yard, handing over papers, et cetera.

"Furthermore, I can't see around trucks in front of me without pulling halfway into another lane."

Elon claimed in his speech that is was easier to see all 4 corners in his semi than competitors, so it's obviously something they're thinking about.

Honda thought t was a good idea to replace volume knobs with capacitive touch strips.

People hated it so much they’ve had to add the knobs back.

A car maker ‘thinking about something’ doesn’t mean the solution will be good. There are tons of infotainment systems to serve as counterexamples.

And let’s not forget GM’s ‘easier’ gear shift that caused people to get run over because you couldn’t tell if it was in park.

Skepticism is always warranted for new things, but also remember this is the guy who made EVs viable and cool and luxurious, as well as they guy who made reusable rockets a thing, to put it mildly. Elon Musk has a brilliant mind for solving engineering problems that others think are too daunting to even approach. "Seeing around corners" is something fairly simple, especially when you consider Tesla vehicles have myriad sensors for detecting objects/vehicles on the road, even through other vehicles. Teslas can stop when the car ahead of them stops--surely they can provide a driver with a (near) real-time view of where cars are in another lane, ahead and behind the vehicle.

but then look at the gull wing doors! I think at this point we can only wait and see.

That would actually be the monostable ZF shifter used in some FCA cars.

You’re right. Wrong brand. It was also FCA who made the weird dial shifter a few years ago that was confishong to use and easy to confuse with the volume knob too?

I had ‘fun’ driving a Chrysler 200 with one of those.

not sure how that's relevant to seeing down the road around the truck in front of you.

Farm equipment still has to be driven with <6" of clearance on either side so it doesn't drive onto produce rows. Semi trucks aren't passing people going 70 mph faster than they are, and vehicles to the side and rear won't be going at a bigger speed difference than 15 mph. You only have to worry about things coming at you at 70 mph from the front.

It's still an absurd comparison. I'd venture to guess that the average combine is driven in a straight line at a set speed at least 95% of the time it's being used.

The worry for combine drivers is that last 1% of the time when they have to get that big machine (with wheels on it takes a lane and a half). Then running at 15mph they have to get from field to field. There is a big problem with other drivers (cars) seeing it from well back and not realizing it is going slow until it is too late and they rear-end it. Or they realize it is going slow in time, but because it is slow they decide to pass on a blind corner (or perhaps pass just as the combine passes a mailbox and so it cannot move over).

Yes rural roads do not have a lot of traffic, but they are not empty.

Could do mirrors that can retract electronically.

Was it stated that there would be no mirrors? It's common for concept/renderings of autos be mirrorless because it looks nice. The mirrors always get introduced at a later stage. Even the new roadster renderings have no mirrors, but surely the will at production.

https://www.tesla.com/roadster/

A lot of those problems seem to revolve around line of sight issues, which won’t matter because the thing will be loaded with sensors.

And one additional thought: When evaluating a potentially disruptive technology, the opinion of veterans is important and may indicate that they won’t ever adopt the new technology in massive numbers. However veterans by definition have a lot of hard earned experience, which means they won’t see the benefits of automation and ai assistance, while people new to the field will have the benefit of not having to gain that experience to be productive. Which means that the old tech will age out as veterans retire.

Sensors are absolutely nice to have, but depending on them is a completely different story.

We absolutely depend on sensors for many things, e.g. airplanes. Why not here?

We absolutely do NOT depend on sensors when backing up airplanes. We have human beings walking the plane out, with those funny lights held up to let the pilot know if they are clear or not. Even in cars with rear-view cameras, you still need to check what is going on in the area outside of the camera view. Yes, sensors help, and I'm glad we have them. But I don't trust the safety of myself or those around me to those sensors. I double-check.

Fair enough, I was thinking solely in flight, but I stand corrected. Thanks

That's a pretty false equivalency in my opinion. Airplanes can still "operate" without some of the sensors, and have a much higher maintenance standard.

As the other comment suggested, Tesla doesn't have the greatest track record with these sensors, and putting 16.5 ton driving force behind that could have very lethal consequences.

I just wish Tesla was held to a higher standard of scrutiny.

Good point, and I agree about scrutiny.

Eyes are sensors too. They've just been historically better than man made sensors. At some point we'll be able to collect enough data using man made sensors and then turn it into a clear enough picture using computers that the benefit of having a human making decisions will be limited at best.

It doesn't even have to be as good as a human could be; it just has to be good enough given the problem space and the economic forces involved.

Everything we do depends on sensors, our eyes, ears, nose, etc... they are all sensors.

I can't imagine not depending upon them, nor how to design something without a way to sense to be able to move through the world.

What viable alternative is there if not to use sensors to understand the world?

Did you read either comment? I was referring to removing the human element from driving (aka. Line of sight concerns)

> Did you read either comment?

Yes I did, your comment would have been improved with specificity in the first place.

Within the context of the conversation, it was very specific. You are just being argumentative in my opinion.

Fair enough, but even in context it was rather unclear at first read.

But on the other hand, Tesla already has a few notable sensor failures, the biggest one being the fatal crash when the sensors failed to see a semi truck at 90 degrees to path of travel (or cancelled it out as noise, etc.)

Obviously things improve with time, but Tesla has never convinced me of having a robust pipeline for validating their sensors/AI projects. Everything they release feels like someone said "well, it's good enough for how much we're willing to spend." Notably, their refusal to evaluate LIDAR when every other company is banking on it.

You don’t think Tesla evaluated LIDAR at all? I’m sure they did preliminary tests to confirm or deny what they could and couldn’t do with it.

Considering every car produced by Tesla in the last two years has Autopilot, I think their pipeline is pretty great at gathering data for every car and performing tests against the data they gather.

LIDAR sensors are expensive. I wouldn’t be surprised if Tesla stopped evaluating them right there and thought, looking at the price of regular camera sensors, “Well, humans have two RGB cameras in their head and seem to do just fine. We can replicate that!”.

As a counterpoint, I personally have had a few notable sensor failures, culminating in an incident when I was 17 that convinced me that I was not safe to operate motor vehicles. No injuries or damage, but I still don't own a car, and my decision is reaffirmed every time I catch a ride from someone and try to follow along to see how it goes.

You may have experienced fewer sensor failures than you've seen Teslas having in the news. But... you're exactly one person, and you're hearing the news about every "sensor failure" that Tesla has had in the entire world. And beyond that, a sensor failure is defined as an event that you don't experience. How many times have you only avoided death by pure chance? Tesla's "notable sensor failures" are literally an every-day occurrence for the mark-one eyeball. The semi crash you're talking about? Humans do exactly that with such regularity that the EU mandates guard rails under the beds of semi trailers.

The difference is that, for Tesla's "sensor failures", obviously things improve with time. You think that humans do the same? Our cars are better, signage and visibility technology are better, but I guarantee you that today's drivers are, in aggregate, the same dumb-ass teenagers that they have been for the last hundred years. And that's why I support Tesla. We can argue forever about whether they're safer than humans today, because driving is such a personal thing that nobody is willing to disentangle their anecdotes from data, but it is obvious that automated vehicles will be safer than humans in the near future. Possibly the very near future.

How'd you figure out Tesla refused to evaluate LIDAR? And if they did, what's the model S with LIDAR that's been spotted near Tesla's Palo Alto HQ doing?

https://www.teslarati.com/tesla-lidar-sensors-spotted-testin...

Sensors can also introduce problems: http://cap.ece.gatech.edu/papers/1569927381.pdf

You may also make the LIDAR think something is in front of the vehicle (when that's not true) and maybe it'll slam on the breaks? Or if you fool the sonar to make it think there's nothing in the way when there actually is, etc.

Lot's of potential here. So in addition to solving problems, sensors introduce many problems and we don't yet fully understand the extent of the problems that have been introduced.

https://en.wikipedia.org/wiki/Shoshin

I'm curious how stuff like a rear view camera is going to work since it seems even more essential now given the seating position. Do truckers now have to install/re-install a camera on trailers when they pick them up? Is it wireless, so that means batteries?

A mini rant on this site - I just got uMatrix installed and the amount of scripts on this page is overwhelming. In spite of enabling most CDN related stuff, the article doesn't show up. Used the "reading mode" and somehow the article shows up.

> I just got uMatrix installed

Sometimes you need to reload the site without cache with Ctrl+Shift+R. uMatrix can be a bit finicky.

Says a horse rider about a car

I think he has some constructive points, don't know if I would really label him a Luddite...

How many years till will won't need drivers?

Tesla should just kill driving and go fully autopilot, it's insane that a human has to drive a truck from point A to B, is just a waste of time and extremely dangerous.

You want to trust an 80k lbs semi in driving rain and sleet on a frozen mountain pass?

There are tons of conditions that even the best autonomous systems are no where near ready for.

It’s going to be a while.

Yes I trust more computer calculations than humans, way more. It will take a while, but It will happen. Anyway, I think there will be better ways to transport than trucks and cars.

Yes I trust more computer calculations than humans, way more

So you're saying you trust code written by humans way more than humans? That doesn't make any sense.

As a developer I was about to give you an elaborated reply but I'm gonna simplify it.

Do you trust more your calculator when you have to perform simple operations like sums , multiplication or square root , or you trust your self with a paper and pen more?

The calculator is a board with written code by humans. Then please, do you think Space X does all their calculations by hand or base on computer simulations using well known and tested algorithms

Do you trust more your calculator when you have to perform simple operations like sums , multiplication or square root

I trust the calculator sure, but I try to always understand enough about the problem at hand to know when the numbers don't make sense - bad data, keying error, edge case, numeric overflow, wrong units, yadda yadda. Punching numbers in and blindly writing down what you get out is a recipe for disaster. When I did engineering it wasn't unusual for people to do the same calculations on an HP and a TI just to make sure - and bugs in calculators weren't unheard of.

base on computer simulations using well known and tested algorithms

SpaceX same as everyone else who actually does this, combines simulations and actual experiments. Noone flies anything built off simulations alone, and I can't see that happening for a looooong time. And even then, things still sometimes go wrong. Computers are not a magic silver bullet, they are just a tool like a spanner or a hammer.

Didn’t NASA land a robot in mars without testing just base on simulations and law of physics , mainly because they couldn’t do in other way. Of course computers are a tool and not a silver bullet , but I trust more the computational power of a computer and the path taken base on algorithms in dangerous situations than a fragile human being

citation needed

Instead of designing cities (the world) base on cars , we will design cities base on the next transportation method, which in my opinion will be autonomous. we won't own cars, for what do you want a car? you won't need it. I think is obvious it just needs someone to push the rest, will be tesla? I don't know, but the car (truck) economy is not sustainable in my opinion.

Can you cite one case where a loaded semi is driven on an icy road? Or a company that promises to achieve that within 5 years?

5 years seems like... short time, as a reminder, 16 years ago the best video game system was Nintendo 64, in 2006 mobile phone were anything near what we have today, Tesla autopilot was first offered in 2014, 3 years ago. I wouldn’t be surprised if by 2019 full autopilot is a everyday thing

Yes.

Because when the autonomous systems are trained on how to handle adverse conditions, its the sum of the knowledge of all machines that build how to handle these conditions. Whereas each human has to learn it individually. And given how bad people drive even with rainstorms, I'd put the money on the computers.

And frankly, we can also discuss if "driving in rain/sleet on frozen mountain pass" is even a good idea. Because computers can also provide data like "Estimated chance of wreck is X%", and decisions can be made using that.

Tl;Dr. One machine learns and can share with all nearly instantaneously. 1 human learns and that's it.

I couldn't agree more

In that case program to accept loss of vehicle/loss of cargo as the failure mode if recovery is not possible, assuming there is no other road directly beneath it which it would land on.

> And given how bad people drive even with rainstorms, I'd put the money on the computers.

In the long term, yes, definitely. And I am looking forward to the day.

But it is going to be a while, and it is going to be a while longer until all the politicians and insurance companies are convinced of that, too.

I can easily see insurance companies being convinced of that a lot sooner. Why, you might ask?

An "auto" will have less incidence of wrecks and drive provably safer. And the "auto" will also record everything as proof, so if there is a wreck involved, it's a simple "download and review the footage". Guilt is now easy to prove as well as trivial to exonerate.

And all of those things will bring the cost of insurance down for the insurance companies. And many of them will then lower their rates appropriately given very high standard of proof they provide.

After all, insurance is heavily based in the sciences and mathematics. If they weren't, insurance companies wouldn't last very long at all.

Now, I have no clue on how to get politicians to understand science. I'm not sure if that's even possible, as science can easily go against an ideology - and thusly their political stance.

I think that your thinking is too shortsighted. Autonomous driving, if it's going to be as rosy as you and I both think, is going to be what kills much of the auto insurance industry. What we have now are more serious and more frequent accidents compared to the ideal future, yielding higher premiums. Higher premiums --> higher cashflow --> higher profitability for insurance companies. Sure, they want to see fewer accidents - but if it's to an extreme to the point where competitive forces dramatically bring premiums down, the insurance business won't be so great. I see insurance companies lobbying very strongly against autonomous driving in the near future.

Interesting thought.

To be honest, once autonomous driving systems become so good they are provably and reliably better than humans, insurance companies will probably charge a premium for the privilege of driving your car yourself.

I am just not sure when autopilots will become that good. OTOH, I did not expect to see self-driving cars arrive during my lifetime, and now it looks very strongly like I will. ;-)

> Now, I have no clue on how to get politicians to understand science.

Given the kind of political influence car manufactures have (at least here in Germany), I think that will not be too much of a problem once the demand is there.

Perhaps it might be a good rule of thumb that no human be allowed to drive a 40 ton semi in conditions under which we don't trust an autonomous system.

No human is ready for driving anything on any road. I do it often, but I know I'm taking a risk and I don't like it. I know a lot of tricks that help, but the fact is it is not possible humanly to do this task to do this task. Any icy road just makes it worse.

You say Tesla should "just" implement self-driving trucks, then in the next reply you say "it will take a while". For goodness sake, I wish self-driving/Tesla groupies would think before they spout this glib time-wasting stuff that they wont even stand behind after one objecting reply. Debate honestly.

It will take a while, as I said, because people won't trust it. then you need sometime for humans to see it as normal, and comprehend than a computer is better than them at driving, and is safer for them if they don't drive.

If people get mad at uber for killing taxi, imagine how mad they will be when uber will kill taxi drivers, probably in less than 5 years. Now imagine Tesla telling us, you know what? you don't need Truck drivers anymore, which we don't. Then the rage would be ...

Las Vegas recently debuted an autonomous shuttle. It experienced its first accident within two hours.

The road to trust could be long and bumpy.

It will, but how many accidents per minute are there by drivers in the world? And we don’t say, puff I don’t trust drivers. Somehow we accept dying by human errors better than by a computer, somehow I always hear, look the computer failed , like if you in that situation as human wouldn’t have done it. I don’t know, I embrace the future , if compouters tomorrow are able to code and My job as a engenieer becomes obsolete, welcome.

This is not in any way realistic. Autopilot failed to notice a semi trailer at a 90 degree angle barely over a year ago. Real world highway conditions have a ton of edge cases, inclement weather, etc etc that automation is simply not ready for. Eventually, we will be there, Tesla is doing great work, but full automation isn't happening anytime soon and until that point, the feedback of those who are actually doing the job is extremely valuable if Tesla wants to make a competitive product. Fanboy screeching about how Tesla can do no wrong is not helpful.

If we ask for 0 margin error , well It won’t happen. If we ask to have less accidents than humans, we are already there.

Without having tried the thing for a few weeks? Not really.

Reply on news.ycombinator.com