Driverless Cars

The thing that baffles me is how we're apparently on the cusp of self-driving cars, but most trains still have human engineers? Surely we should be able to automate the trains first. It looks like Europe and Asia have a decent number of essentially automated metro systems, but in the U.S. it's only monorails and people movers at airports.

Our buses have traffic light control powers here.  Not sure how common that is.

The head of the Toyota Research Institute breaks down a lot of points about level 5 autonomy, or what I've been calling the True Perfect Driverless Car, both in technical challenges and philosophy.

I still expect that for many decades if not the rest of my life the most we will see is ultra-cruise control, not the True Perfect Driverless Car that can drive up my parents' gravel driveway on the edge of the county where there is spotty cell and GPS signal. Much less across half a mile of cow pasture down the fenceline to the fishing pond. Even that level of technology raises some questions. Many states now have strict laws against hands-on cell use (of course people still do it); how is the traffic cop going to know a "driverless car"? How will states test for and issue driver's licenses if people only drive 3% of their in-car time? Will insurance still spike on (disproportionately poor) people who live in places that can't be fully supported by level 4 driverless cars, even though there is no technology to serve them perfectly?

The cop's going to know it's a driverless car because it's a driverless car. When it's here for real they're just going to be 'cars' we're not talking hover conversions of someone's Delorean here.

In any situation of doubt I imagine the burden of proof would be on the driver to use the logs from the vehicle to show "system x" was engaged at the time of the phone call/ticket. Though I also imagine when we're at that point there will also be some sort of terrifying big brother remote query built into cars for the cop to just check system status/destination/blood type etc by scanning the license plate transponder whatever.

Ooh, this is an interesting question with several interesting answers:

1. A lot of it deals with institutional culture, which is just very conservative in railway/transit world. Risk management is done differently in railways than in road transport or aerospace because they held onto hard accident rates longer, rather than thinking about accidents probabilistically. I’m not sure if this is still the case but it’s something that helped sink a French personal rapid transit system, ARAMIS, and slowed the adoption of automated systems in transit in Europe and the US. There’s also a history of thinking trains are safer if there’s a driver, just in case—in many cases making to just having a driver (as opposed to have a conductor or conductors on each train too) was very difficult.

2. Railway automation tends to work via signaling systems rather than at the vehicle level—even for passenger-safety stuff on automated systems it’s relayed via a system on the platform (platform screen doors or laser systems). This means to automate a system you have to redo all the signaling, which is very expensive and hard to recoup relative to just continuing to use drivers—the only time legacy systems tend to get automated is when they also need to have a bunch of other upgrades as well, such as increasing the number of trains as on the Paris Métro or possibly some NYC Subway lines (I think the L) in the future.

3. Given that automation is a signaling thing, if a system has level crossings it’s generally considered not a good candidate for automation. Since level crossings are less expensive than tunnels or bridges everywhere a some systems, even if they’re entirely grade-separated, don’t opt for automation for the sake of keeping future expansion options open. I expect this to change, though—I think Copenhagen’s expecting to automate its suburban trains at some point.

4. Although generally automation leads to higher capacity (40 trains/hour is a pretty typical automated train throughput as opposed to 24 for not-automated), there are cases where the old fashioned hand-brain system does better than modern signaling. The Tremont Street subway of Boston’s Green Line, for instance, where you have a lot of trains going over four tracks at low speeds with passengers all over the place, would probably lose some capacity if automated.

So the Trump administration will probably take a more cautious approach to self-driving car regulation. This was sent to me with the remark that the Trump administration might be trying to slow the automation of labor more generally, which I think sounds like a stretch given the level of coordination and foresight we’ve seen so far (none).

In any event I do wonder if we’ll see an explicitly luddite (as opposed to simply nostalgic) politics emerge in the US or whether we’re too attached to tinkering and innovation as narratives.

I personally knew old men who cut the seat belts out of their vehicles in protest when mandatory seat belt laws were passed because they were never going to be tangled in those death traps if they had to get out of their vehicle in a hurry like if they drove into a river or something.

Humans are such terrible drivers that even a modicum of automation should result in a lot more than marginal improvements in safety.

What is important to remember here is Moore's law (that computing power doubles for the same price roughly every 18 months to 2 years) is exponential.  Presuming it holds, in 10 years, the computers they will be building will be 32 times as powerful as the ones they are using today.  In 20 years, they will likely be 1024 times as powerful as they are today.  My prediction is somewhere between the two, they are going to be able to not only handle this navigation, but handle it much faster than a person can.  Of course, maybe at some point, Moore's law will stop holding, but people have been predicting the end of it for quite a while, and so far it has not happened.

Pedestrians, cyclists, and other suckers will think twice after a few high-profile examples of how driverless cars handle a trolley problem. If a robot has to make a decision whether to veer into a light-pole at speed to save a group of little old ladies jaywalking (killing the driver) or plow through the old ladies (saving the driver), you better believe the company will only sell cars that protect the driver at all costs. Sure, the companies will try to program for this eventuality, just like we raise children to be moral, but the inevitability of a trolley problem is the whole point.

ganews I woner if it’s also due to Uber just being in a very bad place generally PR-wise at the moment—I think if it were a Volvo-branded test, for instance, it would probably be recognized more as growing pains.

Don't worry, the "driverless" Ubers still have a human behind the wheel. And in Pittsburgh at least, another person in the passenger seat. They're so driverless they have even more people in the car! So it's as good as any taxi if you're into that sort of thing.

I don't see this taking so long as "30 years". You're going to be seeing the first trickle of layoffs from long-distance truck automation within Trump's presidency... and there are WAY more jobs in the trucking industry (plus trucks  stops, repair shops, etc) than in most other secors, across the US. Way more than coal. "Insurance" isn't going to be the problem: go look at what % of jobs in your state are Transportation. if it's not top, it's top 3.

So, within 10 years, the trucking industry is going to be in a bad way-- and when trucks are automated and mass-owned, you know their repir garages are going to be all chains as well. Lots of jobs lost.

On the plus side, two automated cars ar two cars presumably in some kind of bluetooth lockstep. So, no accidents, and higher speeds. And more danger to drive among them without an autopilot.

Yes, once everyone has one it will be great. I look forward to the day of ubiquitous True Perfect Driverless cars. It's the transition that's the problem. People have a lot of money invested in their cars; is the government going to mandate that those investments are now worthless? Is this government going to buy a hundred-to-multi-thousand-dollar upgrade for every dumbcar still on the road? Think of it like guns, which last a long time with proper care: you could stop new sales tomorrow, but there's still a lot out there and will be for decades.

I hope this thread is around in twenty years. I think it's at least as likely to be We Were Promised Jetpacks.