Self Driving Cars

[JagerIV]

There's definitely places were they will be useful: moving from a port to a local distribution center. Though that does bring up an interesting thing:

What self driving vehicles should be able to handle really easily and well is very simple, point to point transport. There, the obvious thing to be automated first would be trains. Yet outside of some relatively nitch things, full automation of trains, which should be a lot simpler than cars, seems to be very slow going.

It could be an element of what I've theorized with trucking, that the very small crews necessary are already a very small share of total costs of train operations, so that the gains from getting rid of the few remaining crew on a train are generally not worth development costs.

But, I do think that might be a good break on thinking on how fast things are actually going to go that (full) truck automation seems unlikely to happen before train automation, which is a much simplier task with much fewer moments that might require human involvement (such as buying and pumping gas).

 

[LordSunhawk]

I have a very simple rule. Until the entire board of directors, chief engineers, programming team, and main investors are willing to get in a self-driving bus with no safety driver and traverse Bolivia's Death Road at normal local bus speed, I am not willing to get in a self-driving car of any type.

 

[Francis Urquhart]

What self driving vehicles should be able to handle really easily and well is very simple, point to point transport. There, the obvious thing to be automated first would be trains. Yet outside of some relatively nitch things, full automation of trains, which should be a lot simpler than cars, seems to be very slow going. It could be an element of what I've theorized with trucking, that the very small crews necessary are already a very small share of total costs of train operations, so that the gains from getting rid of the few remaining crew on a train are generally not worth development costs. But, I do think that might be a good break on thinking on how fast things are actually going to go that (full) truck automation seems unlikely to happen before train automation, which is a much simplier task with much fewer moments that might require human involvement (such as buying and pumping gas).

The Docklands Light Railway in London is driverless and fully automated (despite the name it is a passenger service) but I have noted that the safety margins are much higher and the trains slower than on London crewed train services. It's fun though to sit in the front seat and pretend one is driving it. I note that for later people-moving services, the emphasis seems to have returned to crewed service.

 

[JagerIV]

Modest gains are still gains. Did you remember to take into account how much faster and more efficient self-driving cars are going to be, compared to human driven ones? Time is money after all, and fuel consumption should also be considered; not to mention that, with a completely self-driving car, you'd be free to multitask as you're traveling to your destination.

Well, are they going to be faster and more efficient? Especially "much" more?

Most of any potential efficiency gains in travel are going to be not in the self driving part, but in the computer in the office brute forcing something approaching an optimal rout. I wouldn't be shocked if many trucks already basically do cruise control staying in the same lane for long stretches of their journey...which is basically what the robot will do also.

I just don't see much room for improvement in "driving the speed limit in the same lane" between the existing driver and any plausible self driving program.

This is the issue earlier stated with saftey and insurance: Car driving is already 99.9% safe. Truck driving efficiency is probably already at about, to pull a number out of my hat, 95% efficient. At least given current conditions (speed limits, road design, truck design (so how fast can a 40 ton truck actually take a 90 degree turn to an adequate degree of safety?)

I'm not sure how much room there actually is for improvement in any meaningful sense. Like, the sensor might respond to seeing a clueless pedestrian wandering onto the street in 0.001 seconds instead of 0.1 seconds, But if it takes the breaks 1 second to stop and half a second to hit them, your not making a meaningful improvement for that situation, despite the on paper 10x increase in reaction time.

Thus my contention self driving cars, even when fully implemented, represent a 1-10% marginal improvement over existing tech, across a fairly large range of statistics: because I don't see a whole lot of room left to improve: a hundred years of trucking optimization has already done something like 70-90% of the reasonably possible optimizations.

 

[Lanmandragon]

There's definitely places were they will be useful: moving from a port to a local distribution center. Though that does bring up an interesting thing:

What self driving vehicles should be able to handle really easily and well is very simple, point to point transport. There, the obvious thing to be automated first would be trains. Yet outside of some relatively nitch things, full automation of trains, which should be a lot simpler than cars, seems to be very slow going.

It could be an element of what I've theorized with trucking, that the very small crews necessary are already a very small share of total costs of train operations, so that the gains from getting rid of the few remaining crew on a train are generally not worth development costs.

But, I do think that might be a good break on thinking on how fast things are actually going to go that (full) truck automation seems unlikely to happen before train automation, which is a much simplier task with much fewer moments that might require human involvement (such as buying and pumping gas).

From what I understand not automating trains. Has less to do with ability and more to do with the Union blocking it. I'm going off memory though so take that with a grain of salt.

 

[Deleted member]

From what I understand not automating trains. Has less to do with ability and more to do with the Union blocking it. I'm going off memory though so take that with a grain of salt.

Train operations have some complicated features that make automation of freight trains difficult, even long mainline unit trains have to be carefully managed for slack in the couplings and adjusting cocks for brake pressure. Snapping a coupling on a 200 car freight over Tehachapi Pass could delay fifty trains while a crew went out to repair it.

 

[Francis Urquhart]

From what I understand not automating trains. Has less to do with ability and more to do with the Union blocking it. I'm going off memory though so take that with a grain of salt.

Trains in general, I don't know. Docklands Light Railway and the London Tramways are almost identical in use and environment (and IIRC owned by the same holding company, Transport for London). DLR came first and was fully automated. When Tramways experience led to the automated option being explicitly eliminated from consideration.)

This is really the problem we've had throughout this thread. It's been asserted that self-driving vehicles will be faster, safer, more economical etc etc yet there is not one iota of proof to suggest that is the case. In fact the evidence and experience that these claims are not true is overwhelming.

 

[Deleted member]

Trains in general, I don't know. Docklands Light Railway and the London Tramways are almost identical in use and environment (and IIRC owned by the same holding company, Transport for London). DLR came first and was fully automated. When Tramways experience led to the automated option being explicitly eliminated from consideration.)

This is really the problem we've had throughout this thread. It's been asserted that self-driving vehicles will be faster, safer, more economical etc etc yet there is not one iota of proof to suggest that is the case. In fact the evidence and experience that these claims are not true is overwhelming.

It's the same thing with BART being fully automated and none of the light rail systems in the 50 years since it was built being automated.

 

[JagerIV]

Trains in general, I don't know. Docklands Light Railway and the London Tramways are almost identical in use and environment (and IIRC owned by the same holding company, Transport for London). DLR came first and was fully automated. When Tramways experience led to the automated option being explicitly eliminated from consideration.)

This is really the problem we've had throughout this thread. It's been asserted that self-driving vehicles will be faster, safer, more economical etc etc yet there is not one iota of proof to suggest that is the case. In fact the evidence and experience that these claims are not true is overwhelming.

Yeah. Its definitely not that there is no train automation: if you look at the wiki on it, there's a fair number. The fact that there's not more there though suggests that the positions where it is beneficial are fairly marginal: so even if it might be better, the moving costs might make it not worth it: one of the newer automated subway systems in Nuremberg is interesting in seeing what conditions made it worth looking to convert in the first place:

1) Reducing manpower
2) Improve capacity (running trains closer together)
3) Grant money
4) ideally quicker turn around (trains could be activated from storage on a more rapid basis)

Note that two of those are very situational: to get a gain from improved capacity by running trains closer means you have to already be operating close to the limit of human drivers. Which means this limit is only relevant to very busy lines in the first place: for the Nuremberg system, this was relevant for 2 lines: U1 and U2, with U2 reaching this limit at just rush hour. Now, for the Nuremberg system this is a benefit at some time of the day for about 80% of the track, because it sounds like the tracks really were being run close to capacity.

For anywhere that is not at capacity however, being able to run more trains down the track is no benefit at all, since many are probably already running less than the human limit, due to limited passengers.

And even where it does provide a benifit, its probably not huge either.

For example, theoretically running trains twice as often is a 100% increase. However, what if over capacity was only 150%? In that case, either you've gone from having full train to partially empty trains, or your only going to add one extra train, instead of 2. Efficiency boost goes down to 50%.

Then you have an issue of what is the actual benefit of clearing out people faster? Lets say rush hour lasts an hour. What benefit does the train company get by clearing that out in 1 hour rather than 2? In making people wait one hour less? If it doesn't actually change the number of passengers (say the number of subway passengers was fixed), having 1 train do it in 2 hours is basically identical to doing it with 2 trains in 1 in a pure revenue sense. Maybe worse if you then have to have purchased a second train with all those connected costs. And had to upgrade both of those to being autonomous, along with any upgrades you had to do to the tracks themselves.

So, monetarily (ignoring physic benefits, which are real but harder to quantify), that faster clearing out of rush hour is only directly valuable to the degree having to wait less time increases overall train use. So, say that faster turn around only boosts ridership by 20%. Therefore, you get a 20% boost to rush hour traffic, which might only be 40-60% of overall traffic, depending on local conditions. So, the ability to go twice as fast might only boost revenue by 10% overall, between attracting more customers or being able to get away with charging more per ticket in exchange for less wait time.

Quicker turn around is also a very marginal benefit, because most riding is very predictable. From working in a restaurant (McDonalds) where it gets very busy and very slow, little record tables are very good in fact in being able to ramp up or down the number of employees that we predict we'll need to meet demand, and those predictions line up very close to reality. Like, sometimes you'll have a bus come through at 10 pm and suddenly you'll have something like a lunch rush when you only have a skeleton crew, but the number of times that happened over the couple years I worked there I could count on one hand.

So, once again while the flexibility to respond to unexpected surges like that could be nice, your talking about something like 1% of the time where that happens. And even with on call unmanned trains you can ramp up, how many surges are going to be so long that being able to add another train will actually be worthwhile? That it doesn't clear itself up on its own before you could realistically respond, both recognizing the surge, and getting the train to that surge?

So, you've got an ability there that's useful for maybe 5% of operations, where it provides a marginal improvement, maybe increasing the system efficiency overall by 1-2%.

So, between getting rid of drivers (maybe 10% of operating costs), more prompt service during rush hour and to unexpected surges/falls, you might be looking at a 10-20% actual gain in profitability, before considering any expenses. Actually running more trains is a cost. Keeping trains in such a condition to be able to start them up has a cost. A €612 million project price tag is a huge cost.

Not having your trains ready for 2 years longer than expected is a huge, huge cost:

"Construction of the line started in 2003, with the DT3 units ordered in the same year, and opening of the initial line segment from Maxfeld to Gustav-Adolf-Straße had been scheduled for early 2006 to be operational for the 2006 FIFA World Cup.

Initially it was thought by Siemens and VAG that development, testing and certification of the ATC components could be conducted during those 3 years in parallel to the construction of the line, at first in simulations and, after the first DT3 units had been delivered, on a test track at the Langwasser Depot, and that the new line could enter service immediately after the tunnels and stations were built. However, in 2005 news was published that ATC development was not progressing as planned and that the opening would have to be postponed by one year to late 2006 or early 2007. In fall 2006 the responsible parties had to admit that the ATC system would still not be ready by the already postponed date at the end of 2006 and that the opening of the line would have to be postponed again. At that point, Siemens appointed a new project manager. The new U3 line finally opened on June 14, 2008.

To Siemens this delay is a major embarrassment, since the company hopes to sell this ATC system to other subway operators around the world who wish to gradually convert their existing subway lines to ATC operation, allowing for mixed operations on line segments used by ATC and non-ATC operated trains during interim periods."

Further, according to the wiki on self driving trains, the Nuremberg train system remains the only fully automated system in Germany, despite being in operation for nearly 10 years at this point.

 

[Deleted member]

@JagerIV thank you for that very informed discussion of an automated system implementation.

 

[Francis Urquhart]

What the Captain General says. That was a fascinating and very valuable insight into the situation and I thank you greatly for it. One question I would ask and that is automation allowing trains to be run closer together. I'm not an expert on this by any means but from casual observation of DLR and London Tramways, it seems to me that automated trains are run further apart than manual ones. That may be a artifact of my timing and usage but it is interesting. The only accident I know of that was distance related on the manual trains was when a driver (IIRC in Germany) was playing Dungeon Hunter V while he was supposed to be driving the train and didn't notice the train in front had stopped. That's the sort of dumbness that proximity meters were supposed to solve.

 

[40thousandninjas]

The biggest advantage in self driving cars is the complete removal of humans who need to be paid tens of thousands of dollars a year, plus insurance.

While it is true you need to be paid for developers/engineers/production/QA/mechanics support, software can be replicated ad infinitum and distributed completely without a logistics network. It is also true that automation needs to be added for distribution of goods or a human needs to participate. However the entire point is that there is less human involvement. While previously you might have needed three, you now may only need two. That's tens of thousands of dollars a year in savings.

Keep in mind the costs and risks in hiring plus salary plus insurance. A machine is predictable, repeatable and deterministic.

The advantages easily outweigh the costs associated with it. Software engineers may be expensive but we're also far more economically productive than other workers. Any software improvements are disseminated almost instantaneously and further reduce human involvement.

Sure, people here may say they won't trust a machine to drive them and that's fine, all it means is it would take a generation for the technology to gain societal acceptance. Remember at one point some towns in the United States banned "horseless carriages"

It is not a question of if self driving cars become available and mainstream but when.

 

[Darmani]

Welcome to human psychology; obstinate, paranoid, schizophrenic little buggers, aren't we? The very thing that would make self-driving cars safer and more efficient than human-driven ones, is what makes it difficult to convince people to make the switch; even though doing so would be better for them.

don't we have disabled taxicab services?
to counter the think of the disabled rhetoric

 

[Darmani]

Welcome to human psychology; obstinate, paranoid, schizophrenic little buggers, aren't we? The very thing that would make self-driving cars safer and more efficient than human-driven ones, is what makes it difficult to convince people to make the switch; even though doing so would be better for them.

don't we have disabled taxicab services?
to counter the think of the disabled rhetoric

 

[Deleted member]

A machine is predictable, repeatable and deterministic.

That’s a software engineer talking about hardware all right. Let me assure you as a mechie that a plant is damn near a living thing and you can create a mean time to failure with great accuracy for a single part... But half of what we get breaks the day it’s installed and needs to be coaxed into working over months by an experienced technician anyway.

The advantages easily outweigh the costs associated with it. Software engineers may be expensive but we're also far more economically productive than other workers.

You can’t run an ammonia refrigeration plant which freezes chicken, stores TV dinners and makes ice cream like I can. Have fun eating the money... Please be a little more polite to people outside of SV (Silicon Valley).

 

[JagerIV]

What the Captain General says. That was a fascinating and very valuable insight into the situation and I thank you greatly for it. One question I would ask and that is automation allowing trains to be run closer together. I'm not an expert on this by any means but from casual observation of DLR and London Tramways, it seems to me that automated trains are run further apart than manual ones. That may be a artifact of my timing and usage but it is interesting. The only accident I know of that was distance related on the manual trains was when a driver (IIRC in Germany) was playing Dungeon Hunter V while he was supposed to be driving the train and didn't notice the train in front had stopped. That's the sort of dumbness that proximity meters were supposed to solve.

Well, Nuremberg does seem to be a fairly busy line. Trying to figure out relative busyness by dividing daily rider by km of rout. From the various wikis:

Nuremberg U Bahn: 410k daily riders / 37 km = 11,000 daily riders per station
Dockland Light Railway: 340k / 38 = 8,900
London Underground: 5,000k / 402 = 12,000
New York: 5,500k / 394 = 14,000

So, its about 24% more busy than Dockland, about 91% as busy as the London Underground (Dockland is about 75% as busy).

Of course, if that is how busy things need to be for the automation to make sense, your also talking about something like the top 10-20 systems, not all of them.

 

[Darth Robbhi]

don't we have disabled taxicab services?
to counter the think of the disabled rhetoric

Yes, but. Locally, it's called Para-Transit and/or Dial-a-Ride. We also have lots of mobility assistance on public transit of all kinds of stripes, and then there's the whole host of Lyft, Uber and regular taxicab systems if you're in an area they service.

That assumes you live in an area they operate, of course. A major city has a lot more options than the suburbs or rural areas. They, and frankly all mass transit, lack the major asset of private vehicle ownership and ability to operate it: instant responsiveness and the freedom to go anywhere you want the vehicle can manage. Your ability to travel is a lot more constrained by the service's radius of operation, or fare price. Also, ability to utilize them really depends on the individual's needs. Somebody with epilepsy who cannot legally drive but is otherwise as physically mobile as somebody who doesn't have epilepsy is very different from somebody in a power scooter, or a wheelchair.

You're also seriously disadvantaged during a surge event - if everybody needs to get out of town now, Para-Transit and the like are seriously inadequate. Their capacity is designed for blue sky operations. Many buses only have spaces for one or two wheelchairs, for example. Para-Transit and Dial-a-Ride services are equally space limited, since they operate smaller vehicles. Also, they're usually government assets, so they may be doing other things as directed by emergency services, and so not available to you in an emergency.

So yes, they exist, but they're often seriously limited. If you can't drive, your options are severely constrained. That's why losing or having your driver's license taken away is such a psychological blow to people. A large chunk of your independence lies in that little piece of plastic.

 

[JagerIV]

The biggest advantage in self driving cars is the complete removal of humans who need to be paid tens of thousands of dollars a year, plus insurance.

While it is true you need to be paid for developers/engineers/production/QA/mechanics support, software can be replicated ad infinitum and distributed completely without a logistics network. It is also true that automation needs to be added for distribution of goods or a human needs to participate. However the entire point is that there is less human involvement. While previously you might have needed three, you now may only need two. That's tens of thousands of dollars a year in savings.

Keep in mind the costs and risks in hiring plus salary plus insurance. A machine is predictable, repeatable and deterministic.

The advantages easily outweigh the costs associated with it. Software engineers may be expensive but we're also far more economically productive than other workers. Any software improvements are disseminated almost instantaneously and further reduce human involvement.

Sure, people here may say they won't trust a machine to drive them and that's fine, all it means is it would take a generation for the technology to gain societal acceptance. Remember at one point some towns in the United States banned "horseless carriages"

It is not a question of if self driving cars become available and mainstream but when.

That's the marketing pitch, true, but it also doesn't always work that way. You can say advantages easily outweigh the costs, but without numbers your just basically making things up.

On the manpower side, most places experiences are not like that. For example, my small trucking company that I worked for had about 200 people, of which 6 were dedicated IT people to keep out mainframe working. At a ball park estimate, I'd say maybe 50 of out 200 people were truck drivers. Now, since our company only handled last line distribution, we would not be able to get rid of a single truck driver, since the driver is need to talk to the store owner and help unload. Plus until there's some automated system (or gas station attendants come back into the fore) you need them to refill gas. Anything involving cops would likely also benefit from having a person.

So, at my old company, automation of driving saves basically zero in total number of people we need. What it might allow for is to hire slightly cheaper people to do driving, but that realistically is just a reduction of maybe 20-40%, for something that's all up a relatively small share of costs: Stretching the edges of what I knew about our company, pay to truck drivers may be about 1% of company total costs. I sometimes got the sense we paid more in Cigarette taxes (just the taxes) than we did to the drivers.

So, for what might be half a 0.5% savings, you'd have to sink a lot of money into capital costs (because the trucks and the main company HQ are certainly going to need more equipment), retrain everyone, and disturb the existing supply system for the immense amount of time and effort that will take to implement. Or the management and truckers can put all the effort that they could have put into switching over to the new system, to simply building more customer relationships which can be turned into more businesses, and boost sales by 4%, year over year.

I think people imagine software in the real world applying very differently than it does in real life. For example, I've personally lived through two major software upgrades in my time: once at Mcdonalds switching from physical plastic menus to digital menus, the second time when the trucking company I worked for switched payroll.

At McDonalds, it was about 4-6 months before the TV screens consistently showed the right thing, and for management to learn how to fix them when someone hit it with a broom accidentally and it blue screened. I think one of the screens turned out to simply be broken out of the box, and had to simply be returned.

At the trucking company, switching over to a new payroll program, which worked the same as the old one in about 75% of the way, was a 2 year process to implement, from deciding to implement it to having most of the bugs worked out.

Considering the radically different training that would need to be implemented and massive upfront costs as things are disturbed for possibly 5-6 years of figuring out how it works, it doesn't really make sense to bother with the immense hastle and expense of automation, not when there are so many easier ways to save or make 1-2% more per year. Hell, there's probably several things in purchasing, ordering, and warehouse management that would be far less disruptive to automate, and far, far cheaper. Hell, I implemented some automation in the Accounting department, because I realized there were some things that were being inputting by hand that could be auto implemented by Excel. My reforms and automation in the Accounting department probably saved something like a 500 man hours of work with the initial expenditure of about 50 man hours, plus 1-3 man hours for every person who had to be trained on the new automated spreadsheet system, rather than the old manual input system.

People seem to think implementing these reforms results in a straight savings, when in reality it looks much more like this:

Most "reforms" are going to have to go through a period where things are worse than with the old system: your spending more on equipment, management time and effort have to be focused on this particular project to the exclusion of others, people have to be taken off productive work and onto unproductive work (training), and then they're going to being working at less than hoped productivity for a while as they adjust to things and everyone has to work out teething issues.

Earlier I discussed the Wiki available information on the Nuremberg train automation project. In that one, they decided to automate a new line they were building. Above I've mentioned just how marginal the benefits to automation likely were. But, most notably for this example, automating that one line, which at the time had no other traffic until the automated trains were put in place, took 5 years to implement. 2 years longer than had initially been hoped, with the automation slowing down the opening of the entire line while costing an extra couple hundred million.

So, automating trains, which is a very relatively easy automation problem, cost several hundred million, was expected to take 3 years, and ended up being delayed 2 more years, which probably with a conservative estimate cost something like $400 million in lost ticket sales (assuming about 10% of ridership would have been on the new line and paid about $10 a day). So, overall, the project cost, between direct costs and delays in bringing the line up, about $1 Billion dollars. No wonder its delay scared off several other train systems from automation!

So, given history, automation can be a good option, if you have several years and a billion or so dollars to spend on it. Software, though in this case a lot of hardware as well, simply is not that easy to actually implement in a business. Change is hard, long, and often not worth it. Or at least, there's a lot of other things that are less hard, less long, and less risky than a big change like automating drivers.

When some over 1-20 people have not yet found it worthwhile to move on from Window XP some 20 years after it came out, people implementing something trully radical like self driving cars in less time seems a bit far fetched.

link

 

[Deleted member]

@JagerIV that's an excellent description of some of the issues technology migration can drive. Where I work we had 30% of our employees without e-mail for more than a week simply migrating from a corporate Outlook solution to cloud-based Outlook 365. The three 12-hr days in a row to try and finish fine-tuning a SCADA system to control some new fans to meet a regulatory deadline was one of my more epic experiences at work, and it took us about a year to get new two-stage freezer boxes working vaguely reliably, including parts failing within weeks of them coming on-line (fan motors, touchscreen controllers, etc).

 

[Francis Urquhart]

including parts failing within weeks of them coming on-line (fan motors, touchscreen controllers, etc).

This doesn't surprise me at all. The US Navy is going away from touchscreen controllers and going back to previous technology in its next generation battle management system. Touchscreens worked fine in laboratories but failed in the bridge environment, thus illustrating the second law. "Just because its possible doesn't mean its practical."