5 ms·

You started strong -- upgrading the signaling system is a great idea. But tiny cars making point to point visits...really? The fact is, autonomous vehicle syst

You started strong -- upgrading the signaling system is a great idea. But tiny cars making point to point visits...really? The fact is, autonomous vehicle systems cannot solve traffic, due to inherent throughput limitations. If you waved your magic wand and made every vehicle on the road an AV, the only thing it really solves is parking. The roads would still clog. This would also apply to doing it in subway tunnels, with the added downside of now your subways are just as jammed as the surface roads.

jonahbenton 3y ago
Thanks for the response. Traffic/contention is simply not an issue on the NYC track system. Especially in the outer boroughs. The issue is the opposite- the existing infrastructure is deeply, criminally underutilized. It is in fact underutilized for the massive 10 car people movers- hence the goal of the numerous multi-decade signal and sensor upgrade projects is to enable better utilization by long trains. But even doubling utilization by long trains- still a decade or more away at best- leaves vast swaths of the tracks completely empty nearly all day. It is this capacity that should be utilized by new tiny autonomous cars with their own scheduling capability and that are suited to provide a complementary utility. I would agree that there are way (way) too many cars on surface roads. After the lack of use of the subway tracks the next worst allocation/utilization of space in NYC is free street parking. Although I own a car I think private cars should be banned in Manhattan (yes, I am deeply familiar with the various policy arguments, my opinion is what it is) and better point to point autonomous utilization of underground capacity would go a long way towards that vision.
- warning26 3y ago
  It seems to me that the problem of underutilization could be solved by simply increasing the frequency of trains. (Shorter trains are also an option of course.) Your hybrid system is interesting, but consider the following: Where do AVs not on active rides go? What about if one breaks down? How can you ensure that if a rider at a far-flung station summons an AV, there isn’t an extremely long wait? How do you ensure that the AVs don’t interfere with regular trains? (Especially considering that even now there are struggles with signaling)
  - jonahbenton 3y ago
    Can't increase the frequency of trains with the state of the signal system, the improvement of which should be a goal, of course. But it has been a goal since the city's capital outlook started turning positive 30+ years ago, with no tangible customer-visible accomplishments (unlike, for instance, the 2nd avenue line that finally opened after 50 years of off and on work). Most modern systems can run trains literally very close to one another; in NYC, some lines still require a half mile of spacing (!). Anyway. Where does a fleet of, say, 1000, to start, tiny cars go, and what about passengers in far flung stations? This is actually one of the main use cases. Right now it sucks to live at the end of any line that terminates far out in a borough. At night you can be waiting 20, 30, even 60 minutes. It's terrible. Running long trains all over the system at off hours is just not a thing. So, in theory, as autonomous entities, cars go wherever they want. Like elevator cars, they can strategically position themselves around the system and move and maneuver out of the way when they need to, and be ready to respond in single digit minutes for passengers at the ends of lines at off hours. The economic value of making the edges of the system actually viable for commutes would be one of the enormous benefits. What happens when one breaks down? In practice, I am sure this is a first order concern, but there are autonomous full length trains running in many systems around the world. This is a solveable problem. Even judging by the NYC system, the ability for an electric train to move is last to fail. The mechanicals are dirt simple. When trains in NYC need to be "taken out of service", passengers get off, and the train moves out of the station. Extremely, extremely rare for a train to actually be physically unable to move. In terms of interfering with existing trains- well, I see this also as a problem we have solved. Current train locations are generally well known, as evidenced by the ability for the MTA to provide reasonably accurate (at seconds level resolution) estimates for the arrival of trains in stations. The whole system is mapped. There are lots of junctions and interchanges. Routing and so forth on such a landscape around an existing fleet of known position and trajectory large trains is a solved problem. Self driving cars have made great strides in things like real time collision avoidance and so forth. This problem is enormously simpler on train tracks. There occasionally objects and people on the tracks, but vastly fewer than up on surface roads. Critically there is no weather, and very little light variation. The biggest obvious problem to me is networking. Surface fleets of course talk to satellites. We can't do that down in the tunnels. And the existing carrier networks- with beacons in many stations themselves- offer poor quality service. To my mind the clever solution here is a mesh, and part of the job of the autonomous cars would be to move around to dynamically maintain the mesh with each other. But I would guess there are lots of edge cases given the varied physical architecture of the system and therefore lots of failure modes. So would take some study. This also is an area where there have been decades of more conventional attempts. The need for networking in stations has been evident since the 1990s. MTA has tried many times to outfit the stations with coverage, with marginal success. The most recent networking capital project is planned well into the 2030s. I remember having conversations with folks in the 1990s looking at meshes that could be deployed on existing train cars themselves, one of those ideas maybe a little too radical at the time, but things have come a long way since then. Of course all project ideas like this are fanciful, but I am just surprised that an idea like this doesn't get even some airing. The benefits of targeting a service complementary to the current offering using the same infrastructure would be enormous and in conceptual terms at least the problems ahead of a planning stage should all be solveable.

jonahbenton 3y ago

Thanks for the response. Traffic/contention is simply not an issue on the NYC track system. Especially in the outer boroughs. The issue is the opposite- the existing infrastructure is deeply, criminally underutilized. It is in fact underutilized for the massive 10 car people movers- hence the goal of the numerous multi-decade signal and sensor upgrade projects is to enable better utilization by long trains. But even doubling utilization by long trains- still a decade or more away at best- leaves vast swaths of the tracks completely empty nearly all day. It is this capacity that should be utilized by new tiny autonomous cars with their own scheduling capability and that are suited to provide a complementary utility. I would agree that there are way (way) too many cars on surface roads. After the lack of use of the subway tracks the next worst allocation/utilization of space in NYC is free street parking. Although I own a car I think private cars should be banned in Manhattan (yes, I am deeply familiar with the various policy arguments, my opinion is what it is) and better point to point autonomous utilization of underground capacity would go a long way towards that vision.

warning26 3y ago
It seems to me that the problem of underutilization could be solved by simply increasing the frequency of trains. (Shorter trains are also an option of course.) Your hybrid system is interesting, but consider the following: Where do AVs not on active rides go? What about if one breaks down? How can you ensure that if a rider at a far-flung station summons an AV, there isn’t an extremely long wait? How do you ensure that the AVs don’t interfere with regular trains? (Especially considering that even now there are struggles with signaling)
- jonahbenton 3y ago
  Can't increase the frequency of trains with the state of the signal system, the improvement of which should be a goal, of course. But it has been a goal since the city's capital outlook started turning positive 30+ years ago, with no tangible customer-visible accomplishments (unlike, for instance, the 2nd avenue line that finally opened after 50 years of off and on work). Most modern systems can run trains literally very close to one another; in NYC, some lines still require a half mile of spacing (!). Anyway. Where does a fleet of, say, 1000, to start, tiny cars go, and what about passengers in far flung stations? This is actually one of the main use cases. Right now it sucks to live at the end of any line that terminates far out in a borough. At night you can be waiting 20, 30, even 60 minutes. It's terrible. Running long trains all over the system at off hours is just not a thing. So, in theory, as autonomous entities, cars go wherever they want. Like elevator cars, they can strategically position themselves around the system and move and maneuver out of the way when they need to, and be ready to respond in single digit minutes for passengers at the ends of lines at off hours. The economic value of making the edges of the system actually viable for commutes would be one of the enormous benefits. What happens when one breaks down? In practice, I am sure this is a first order concern, but there are autonomous full length trains running in many systems around the world. This is a solveable problem. Even judging by the NYC system, the ability for an electric train to move is last to fail. The mechanicals are dirt simple. When trains in NYC need to be "taken out of service", passengers get off, and the train moves out of the station. Extremely, extremely rare for a train to actually be physically unable to move. In terms of interfering with existing trains- well, I see this also as a problem we have solved. Current train locations are generally well known, as evidenced by the ability for the MTA to provide reasonably accurate (at seconds level resolution) estimates for the arrival of trains in stations. The whole system is mapped. There are lots of junctions and interchanges. Routing and so forth on such a landscape around an existing fleet of known position and trajectory large trains is a solved problem. Self driving cars have made great strides in things like real time collision avoidance and so forth. This problem is enormously simpler on train tracks. There occasionally objects and people on the tracks, but vastly fewer than up on surface roads. Critically there is no weather, and very little light variation. The biggest obvious problem to me is networking. Surface fleets of course talk to satellites. We can't do that down in the tunnels. And the existing carrier networks- with beacons in many stations themselves- offer poor quality service. To my mind the clever solution here is a mesh, and part of the job of the autonomous cars would be to move around to dynamically maintain the mesh with each other. But I would guess there are lots of edge cases given the varied physical architecture of the system and therefore lots of failure modes. So would take some study. This also is an area where there have been decades of more conventional attempts. The need for networking in stations has been evident since the 1990s. MTA has tried many times to outfit the stations with coverage, with marginal success. The most recent networking capital project is planned well into the 2030s. I remember having conversations with folks in the 1990s looking at meshes that could be deployed on existing train cars themselves, one of those ideas maybe a little too radical at the time, but things have come a long way since then. Of course all project ideas like this are fanciful, but I am just surprised that an idea like this doesn't get even some airing. The benefits of targeting a service complementary to the current offering using the same infrastructure would be enormous and in conceptual terms at least the problems ahead of a planning stage should all be solveable.