Connected/Automated vehicles and infrastructure in Boston

[tysmith95]

Still I think we are ten years into a twenty year transportation revolution. I wouldn't expect my next car to be autonomous, but maybe the one after that.

Self driving cars may end car ownership for most people, especially in the city. Why have cars sitting for 22-23 hours a day?

 

[George_Apley]

Why would hills and bridges pose any difficulty for self-driving cars? Bridges are basically just perfectly straight divided highways.

via Business Insider

During a recent test drive crossing the Allegheny River, Uber's driverless Volvo signaled to the driver on-board to take the wheel. It was only for a few seconds, but it shows that the driverless cars aren't ready to handle bridges without supervision.

That's because Uber has meticulously mapped roads so that the driverless car can compare what it's seeing with what is supposed to be there, helping it avoid objects and pedestrians.

Because bridges don't have many environmental cues like surrounding buildings, it's hard for the Uber car to figure out where it is, Krikorian told Bloomberg. GPS helps the car position itself, but not to the accuracy Uber wants.

John Dolan, principle systems scientist at Carnegie Mellon's Robotics Institute, told Business Insider that bridges can pose other problems for driverless cars, as well.

"You have a lot of infrastructure on the bridge above the level of the car that we as humans take into account," he said. "But when you sense those things with a sensor that doesn’t have the domain knowledge that we do ... you could imagine that the girders coming up from the side of the bridge and that kind of thing would be disturbing or possibly confusing."

 

[tysmith95]

^thats an easy fix. Just install some sort of reflector or sensor that the car can see.

The best way for self driving cars to become safe would be to work with cities to create infrastructure that makes it easier for the cars.

 

[George_Apley]

^thats an easy fix. Just install some sort of reflector or sensor that the car can see.

The best way for self driving cars to become safe would be to work with road builders to create infrastructure that makes it easier for the cars.

No one's saying it's insurmountable. It's a development challenge. Hence the pilot in a city that's replete with such challenges.

 

[CSTH]

^thats an easy fix. Just install some sort of reflector or sensor that the car can see.

The best way for self driving cars to become safe would be to work with road builders to create infrastructure that makes it easier for the cars.

Yeah that's fair.

After all as we all know our roads didn't really become available to conventional motor vehicles a century or so ago until all the other users had been evicted by the public authorities working on behalf of auto manufacturers (or simply run down by motorists) and the cities defaced by a web of garish paint and intrusive signage ... there's no reason to expect that we would welcome the advent of the self-driving era without a similarly huge public subsidy, concentration of public and economic power, and associated civic vandalism...

I for one do not welcome the robots of our new overlords....

 

[stefal]

Re: bridges/hills,
I don't know how many of you are familiar with it, but Tesla has a system in place called "fleet learning" where, if the car doesn't know what to do, it lets the driver take over, while simultaneously collecting tons of data. Then it sends the data to every other Tesla so the car knows exactly what to do in that particular instance. IMO, that's the easiest, most efficient and least costly solution.

 

[whighlander]

Re: bridges/hills,
I don't know how many of you are familiar with it, but Tesla has a system in place called "fleet learning" where, if the car doesn't know what to do, it lets the driver take over, while simultaneously collecting tons of data. Then it sends the data to every other Tesla so the car knows exactly what to do in that particular instance. IMO, that's the easiest, most efficient and least costly solution.

Stefalrchitecht -- tht might seem prescient -- but its actually Hubristic not Heuristic

The number of possible bizarre situations is immense and the probability that most will recur in any reasonable time is very small

Yet its precisely these bizarre situations which cause both humans and robots problems -- robots remember everything but don't do very well in associating near-to's while humans don't remember much of anything perfectly, but do much better at associating things that seem somehow similar -- the "Je ne sais quoi" effect

So now we come to the crux of the self-driving problem -- the interaction of cars with softer things in an unruly place such as Boston such as people and bicycles, and in some other cities such as Bangaluru cows, or in Taipei clouds of motorbikes. Robot cars hitting other cars, robot or not at slow speeds can be tolerated [just grist for the body shops] but robot cars running over or into bicycles or pedestrians can't be tolerated.

These vulnerable other things can approach the car from nearly any angle and often might be screened by near-by vehicles up to the last possible instant [e.g. the pedestrian darting from between a pair of tall vehicles whilst crossing the street at an angle; or the bicyclist cutting through an intersection against the light and suddenly changing lanes or turning].

It's going to be a while before the self driving car is qualified to handle these kinds of situations in a dense urban environment complete with a non-rectilinear street network -- aka Boston.

On the other hand there is probably no good reason why very soon self driving cars, or human driven cars with autopilots shouldn't have access to their own highly efficient highway lanes with substantially reduced spacing.

 

[fattony]

Re: bridges/hills,
I don't know how many of you are familiar with it, but Tesla has a system in place called "fleet learning" where, if the car doesn't know what to do, it lets the driver take over, while simultaneously collecting tons of data. Then it sends the data to every other Tesla so the car knows exactly what to do in that particular instance. IMO, that's the easiest, most efficient and least costly solution.

Brilliant. Machine learning is essential to this type of problem. Neural nets (in hardware and software) are making great progress. Hopefully Tesla and others get access to the best new developments.

 

[stefal]

Stefalrchitecht -- tht might seem prescient -- but its actually Hubristic not Heuristic

The number of possible bizarre situations is immense and the probability that most will recur in any reasonable time is very small

So now we come to the crux of the self-driving problem -- the interaction of cars with softer things in an unruly place such as Boston such as people and bicycles, and in some other cities such as Bangaluru cows, or in Taipei clouds of motorbikes.

It's going to be a while before the self driving car is qualified to handle these kinds of situations in a dense urban environment complete with a non-rectilinear street network -- aka Boston.

The fleet-learning that I was referring to is mostly for complex or tight turns and parts of the road where there's a bridge and the car doesn't know that its a bridge, not for pedestrians. These situations recur whenever "Car B" is driving a specific curve or approaching a specific bridge where "Car A" previously collected the data on how to handle the situation, and "Car A" tells "Car B" what to do.

These cars are also rated level 2 or 3 (depending on how you look at it) out of 4 or 5 on the autonomous driving level, meaning the car still needs the driver to be alert, and Teslas frequently check to make sure your hands are on the wheel, so that you're ready if there's an unexpected pedestrian or if something else pops up.

While they haven't perfected how to deal with pedestrians, they still detect and avoid them. The "autopilot" system on these cars are incredibly complex, and the new radar system they're implementing can still "see" what's in front of it in 0 visibility situations.

We haven't even seen fully autonomous cars yet, so we don't know what to expect of how they perform in the city or with pedestrians. We also haven't seen enough mileage out of the current Teslas, yet. I'm pegging 2020-2022 for the first fully autonomous car, so we'll see then.

 

[elemenoh]

I certainly hope there will be some way for local municipalities to tell these autonomous cars that they are not allowed on local streets. If they can't operate on winding, weird, narrow, 17th-century roads, they shouldn't operate there.

Could you imagine trying to take one of these things down Commercial Street in Provincetown?

It's neat and nifty tech, and the auto-parking and whatnot are clever, but these fully autonomous private vehicles seem as realistic as regular shuttle service to the Moon.

And as an aside, why aren't these features being implemented on public transit? Seems like automated buses, streetcars, and rail would be an easier place to start.

Edit: Just answered that question with the Globe article about Bridj:

George said he’s working on a long-term plan to automate Bridj’s buses so they could navigate city streets without a human at the wheel. “We’re going to start to introduce autonomous vehicles and autonomous delivery devices over the coming months, using Boston as a laboratory,” he said.​

 

[tysmith95]

^Copenhagen's metro system which opened in 2002 is driver-less. Vancouver also has a driver-less rail system.

Buses operate in the same challenging environment as cars. Self driving buses will come about around the same time as self driving cars.

 

[CSTH]

^could probably raise the speed limit for the silver line in the piers tunnel and on the Chelsea segment with some off the shelf technology like tomorrow if we wanted to

 

[Arlington]

Driverless rail systems are almost too numerous to count--not just all the airport ones (ORD, ATL, IAD...) But airport-to-city (OrlyVAL) and full subways (Toulouse, France) and elevated (London Docklands). Closed Bus systems are a natural next, and semi closed like Silverline Waterfront and Gateway would be next.

 

[whighlander]

Driverless rail systems are almost too numerous to count--not just all the airport ones (ORD, ATL, IAD...) But airport-to-city (OrlyVAL) and full subways (Toulouse, France) and elevated (London Docklands). Closed Bus systems are a natural next, and semi closed like Silverline Waterfront and Gateway would be next.

Arlington -- no one is arguing against closed systems -- the Red Line, Orange Line. Blue Line and Green Line to Riverside could easily be driverless. It gets more of a challenge when you have the Green Line in a protected right of way but interacting with catrs, bikes and pedestrians at intersections [e.g. E until Brigham Circle]

However, the "Gap you have to Mind" between the Tube and a Bus in London is wider than the Thames by Parliament -- got on a roll there sorry

If the Silver Line was restricted to the tunnel and the relatively simple grid of streets in the Seaport / Innovation District -- it might be able to be driverless.

However, throw in the trip to Logan and the challenges of dealing with potentially hordes of less than fully vigilant people coming and going at the terminals -- now the current technology will be severely stressed. Take the branch that goes from Dudley into the "Ladder District" -- No way doable for a considerable period of time

 

[Justin7]

these fully autonomous private vehicles seem as realistic as regular shuttle service to the Moon.

Why? There is a lot of money to be made, a lot of lives to be saved, and a lot of very smart people working on problems that while moderately complex are well within the capabilities of already existing technology.

 

[JeffDowntown]

Why? There is a lot of money to be made, a lot of lives to be saved, and a lot of very smart people working on problems that while moderately complex are well within the capabilities of already existing technology.

The existing capabilities only work if you augment the environmental infrastructure, particularly for dense, confusing, multimode locations like Boston/Cambridge.

That augmentation of the environment means spending infrastructure dollars -- something we seem incapable of doing.

 

[JumboBuc]

The existing capabilities only work if you augment the environmental infrastructure, particularly for dense, confusing, multimode locations like Boston/Cambridge.

That augmentation of the environment means spending infrastructure dollars -- something we seem incapable of doing.

The existing capabilities aren't "fully there" today, but they're getting closer every day. First comes "assisted cruise control" (which we already have), then fully automated cruise control (which we are right on the edge of), then fully automated long haul trucking (which would be worth billions), then eventually fully automated everything.

And infrastructure spending is something that we are absolutely capable of doing when it is tied in with private sector initiatives. As long as "there is a lot of money to be made", it will happen.

 

[tysmith95]

Self driving cars should be a bigger priority to the feds then terrorism. Human driven cars kill nearly 40,000 people a year, terrorism in comparison is a miniscule problem.

 

[JumboBuc]

Self driving cars should be a bigger priority to the feds then terrorism. Human driven cars kill nearly 40,000 people a year, terrorism in comparison is a miniscule problem.

But if the enormous amount of time and energy devoted to terrorism tells us anything, it is that government resources are not allocated according to any sort of lives-saved-per-dollar-spent calculation.

 

[stefal]

And infrastructure spending is something that we are absolutely capable of doing when it is tied in with private sector initiatives. As long as "there is a lot of money to be made", it will happen.

But if the enormous amount of time and energy devoted to terrorism tells us anything, it is that government resources are not allocated according to any sort of lives-saved-per-dollar-spent calculation.

Can you or someone else clarify how the private sector is going to pick up some/most of the bill? I don't see how they can make a profit by investing in infrastructure. Seems like the only way we can get infrastructure updates is through infrastructure bills, which are very rare lately, and, like you mentioned, inefficient.