Autonomous vehicle (AV) technology is maturing from what I reported two years ago. While Waymo, Zoox (Amazon), and Tesla compete to replace taxis in the U.S., China (Baidu, Pony.ai, WeRide) is already a bit further along. The rollout has shown that AV technology requires expensive sensors but proved itself unless there is a major disruption.

What about if you install such technology in a bus? Many transit providers in Germany and elsewhere in Europe have partnered with startups such as Holon or Navya or more traditional electric bus manufacturers such as Karsan to test on-demand and fixed-line autonomous bus services. This often happens in partnership with the AV software platform provided by Beep. WeRide is partnering with Yutong for global rollout. These vehicles are usually smaller than traditional buses to serve routes with less ridership. As the buses do not require a driver, they may allow Metro to provide fixed services in low ridership areas where King County currently operates their on-demand Flex service.

The Metro Flex service currently uses traditional minivans with drivers. They have partnered with Via to operate this service. Via already has partnerships with vendors such as Navya and May Mobility. At some point Metro may either offer fixed routes using AV buses in such areas and/or partner with one of these AV providers to operate a driverless on-demand service. Jacksonville Transportation Authority is already testing such AV based service. Holon is already building a factory in Florida to supply them to Jacksonville and the U.S. market in general.

86 Replies to “Autonomous Buses”

  1. I think I could see applying autonomous tech to a railed vehicle like a trolley as a good intermediate step to raise confidence in the tech and troubleshoot a lot of the issues. Being on a fixed guideway removes a significant number of tasks and permits the tech to focus on safety.

    Maybe automate some runs of the FHSC?

    2. It would be easier to automate the streetcar than it would our light rail lines. The trains travel much faster and thus need more time to stop. Given the relative speed of the vehicles, light rail would do a lot more damage. Siemens has been looking into it: https://web.archive.org/web/20220706173318/https://www.mobility.siemens.com/global/en/portfolio/rail/rolling-stock/trams-and-light-rail/autonomous-tram.html.

      Note: That is an old link. So it is possible they stopped the research.

      3. Definitely easier to automate light rail rather than a streetcar.

        The higher speeds make automation *safer* that manual, because an automated vehicle will have faster response times, aka can break faster. That split second faster response is more important the faster the vehicle runs.

      4. AJ, I’m not sure about that but an automated streetcar could be problematic. Will anybody bring up an automated light rail in the webinar today? I’m planning to attend.

      5. So much nonsense and artificial fear is spread about the stopping rate of rail vehicles. The science does not really support this. Though very slightly slower, urban rail vehicles like buses and streetcars stop at a rate of 5.0 mph/s. Buses do only a little better under ideal conditions, but not that much, especially since they so often skid into other vehicles post crash.

  2. Unleashing drone cars that could be programmed with ill intent is a hard “No” for me.

    Given how drone planes are being used, it is hubris to assume good intent for all car programmers.

    1. This is a risk, but you have to balance it against the fact 40k people are killed and 2.4 million people are injured in vehicle crashes every year in the US, largely as a result of driver error. Even if drone cars resulted in only a 50% decrease in driving errors, that would be a huge decrease in deaths, injuries, and property damage.

      1. There are less risky ways to reduce traffic deaths than to remove non-drone cars from the market (which is what your supposition would require, even though a supermajority of cars buyers may be totally uninterested in buying a drone car).

        Safer street design, lower speed limits, right-of-way for other modes (including walking), protected from cars …

        These don’t require a (unlikely) universal shift in individual car buying practices.

        When the drone car industry proclaims their cars to be safer, we have to ask, “Safer for whom?”

  3. Train automation should be a lot simpler than buses that need to weave in and out of traffic. The notion that you need grade separation for train automation is just more of the same misguided 1980s mentality that needs to die off. So many at grade systems including link could easily be automated with AI, camera and sensors systems which don’t need to sleep and don’t require pensions and vacations or call in sick.

    1. We are already getting some of the benefits, with automatic train control. Amtrak had to install it after the failure of several humans on the Point Defiance Bypass. I don’t know if Sounder, Link, the monorail, or the streetcars have any ATC features.

      Deploying more safety technology is the huge safety improvement. Getting rid of one staffer on board is just saving money, a cheap-out we may come to regret.

      1. I agree. We should implement the ATC features even though we have drivers (that is common).

        Having Link be completely automated would be beneficial. But it wouldn’t save nearly the amount of money that automated the buses would. Link carries a huge number of riders per train. The buses (or the streetcars) do not. Just think of how many buses it takes to equal the ridership of Link. You’ve probably got all of the RapidRide and several other (high-performance buses) combined. Now imagine how bus drivers that is compared to how many Link drivers. It isn’t even close. There are way more bus drivers. That means you could run a lot more of the other buses.

        To be clear, I’m all for automating the trains. It makes sense to take advantage of existing technology when we can. But transitioning the buses would be a much bigger change than transitioning Link.

      2. The public isn’t going to embrace drone buses before they see automation work on trains. (This second-order thinking could yield pushback against train automation, FWIW.)

      3. They already see automation work on trains. People also see automation in some cars. I think they will embrace automation on buses as soon as it is feasible.

      6. “Having Link be completely automated would be beneficial. But it wouldn’t save nearly the amount of money that automated the buses would.”

        You are correct on the math of number of drivers, but from a transit perspective it might work the opposite. Vancouver has much better transit use than Seattle, and it is not because they have cheaper or more frequent busses. The fully automated Skytrain provides a very frequent *and fast* backbone that the busses connect to, making the busses useful because you can transfer to Skytrain without a long wait and then have a fast trip to complete your journey. Link has much better vehicle speed than busses because of the huge investment in grade separated ROW for almost the whole system, but frequency is currently not great. Make it so Link trains come every two minutes all day and you could see ridership on the trains and connecting busses soar.

        Note that with the current state of AV technology you really don’t have to choose. If Waymo’s and Zoox’s are driving on busy city streets right now then you can automate the trains, busses, and on-demand services.

      7. The bigger question is with AVs do we need the trains at all. Link at 8tph and 4 car trains is moving 3,040 feet of vehicle per hour. That is only 76 standard busses per hour, or one every 47 seconds. That is nothing, the Lincoln tunnel XBL lane moves a bus every 10 seconds, or 600 per hour. There should be serious questions about if it makes sense to build Link to Everett when you could buy and put a few hundred AV busses on I-5 at a fraction of the price, even if you put money into creating a dedicated lane. I believe it has been shown how Sound Transit Express busses will be faster to Everett or Tacoma than Link because the busses won’t stop on the freeway. If you are running say 180 busses per hour then you have the opportunity to give this kind of service for all stops, not just express busses to downtown. Not every bus would stop at every station, and when a bus pulls in a demand response software system can tell certain passengers to get on based on the destinations, or reschedule where the bus will stop based on who gets on. Modern elevators already do this is large buildings. With that many busses even if only one out of six are stopping at a given station that is still a bus every two minutes, and even if you have to wait a bus or two for one going to your destination that is still only a 4 minute wait at worst, half the waiting time of Link at peak. It could be much less as it doesn’t have to be a fixed “every sixth bus per station” – when a passenger walks in a presses a button saying “I want to go to Northgate” the system can call the next bus going to Northgate that hasn’t made too many stops yet and you get picked up very quickly. Now you have the frequency of a Skytrain with the speed of a highway bus – the best of both worlds.

        Apply the same logic to Rapid Ride. Now you have busses every few minutes, and instead of stopping every half mile or mile at each stop, they stop every 2-3 miles based on demand response, because the many other busses on the road are making the other stops. If Rapid Ride D has the same frequency and speed as BLE would have, is there a point to spending billions to tunnel under the ship canal?

        Apply the same logic to regular bus routes. What would transit usage look like if every bus had the effective speed and frequency as Rapid Ride?

      8. “ A train operator would hop on the train at Mt Baker station (once we build overpass in SODO).”

        I think that it may be very possible to have a virtual driver based in a special simulated reality “room” that mimics a train. It would take more micro cameras to pull it off but it seems fairly easy to create. Of course it would take one room for each train, so there would need to be several created. But once a train switches back to automation, the remote driver can just be expected to push a button or two and they’d be instantly set to drive the next train that needs to transfer to manual operation segment.

        And over time, even that would no longer be needed. AI would eventually replace human divers.

      9. Vancouver has much better transit use than Seattle, and it is not because they have cheaper or more frequent busses. The fully automated Skytrain provides a very frequent *and fast* backbone that the busses connect to, making the busses useful because you can transfer to Skytrain without a long wait and then have a fast trip to complete your journey.

        Yes, but my point is that ST could provide the same level of train service for not that much money. Just hire more drivers. This would cost more money, but not a lot more. The main reason they don’t do that is because ST is focused on expansion, not service. Their focus on headways is almost entirely based on capacity. Trains used to run every 6 minutes down Rainier Valley. Now they run every 8 (at best). Why? They got bigger trains. For that matter, why not just run the trains every 8 minutes, all day long. It can’t be that expensive. Yet they don’t.

        Of course it would be great to automate the trains. I’m fully in favor of it. But I’m saying that as nice as it would be, it pails in comparison to automating the buses simply because we spend a lot more labor on operating the buses.

        As for Vancouver, the key is the integration of the system. Pretty soon the Link trains will be running every 5 minutes midday from Downtown to Lynnwood. This is similar to what SkyTrain offers in Vancouver (3 to 6 midday). Will the north end have a system like Vancouver then? No. Not even close. Our buses just aren’t that good (and we don’t have enough train stations).

      10. I think that it may be very possible to have a virtual driver [for sections like Rainier Valley]

        I agree. I suggested the same thing years ago. The trains would basically be automated on the sections that are already grade-separated. Then, for the sections that aren’t (like Rainier Valley) they operate with remote “driver assist”. This is similar to how automobiles operate but in this case the drivers are remote. They operate the train more or less like it is operated now. So, for a northbound train, the driver starts monitoring things as the train approaches Rainier Beach. They continue to operate the train until it reaches CID. Then they wait for another train. You would need something similar in Bellevue (although the section is much smaller). I’m not gonna bother to do the math but it is clear you would need a lot fewer drivers.

        I’m not sure if anyone does this though. It seems pretty straightforward, but agencies seem focused on full automation. I get why — it would greatly simplify the process. But it seems like we could implement this sort of thing very soon.

        Of course the ideal solution is to completely grade-separate our system. That would provide plenty of other benefits — like more consistent run times and better headways from the south. (The city limits the number of trains on Rainier Valley because of traffic impacts.)

      11. Even if a fully automated train can operate on MLK without a driver present, a virtual driver capability would still be useful.

        I’m thinking about periodic concerns like service disruptions creating overflowing platforms, obstructions near the roadway, non-track problems like signal power outages or malfunctions, nearby street protests that may drift into the trackway, criminal activity or violence in front of a train, and those kinds of things.

        Special situations do occur where real-time human observation is useful if not sometimes essential. Maybe it results in nothing or maybe just a real-time audio warning or information — like how drivers today can blow a horn to warn people positioned too close to the trackway.

      12. Even if a fully automated train can operate on MLK without a driver present, a virtual driver capability would still be useful.

        Yes. Similarly it might make sense to have a special crew stationed in Rainier Valley to deal with occasional snafus (of the sort you mentioned).

      13. “Trains used to run every 6 minutes down Rainier Valley. Now they run every 8 (at best). Why?”

        8 minutes is its policy choice. 6 minutes was a temporary expediency between when the Ride Free Area ended and the full U Link trainset order was ready. ST never promised 6 minutes permanently, and it has never gone back to it. However, it intends to go to 6 minutes in ST3. For annoying reasons we’re supposed to wait 15+ years for this. Both ST and Metro have a tendency of postponing frequency increases until a new Link/RapidRide/Stride line opens, even if the opening gets delayed for several years. Passengers just have to suffer until then.

    2. Train automation should be a lot simpler than buses that need to weave in and out of traffic.

      I don’t think that steering is the tricky part. It is counter-intuitive but a lot of cars right now have hands free highway driving (https://www.ford.com/technology/bluecruise/). Thus you can drive on I-90 for miles — changing lanes, speeding up, slowing down — based on traffic. That is because it only expects cars on the road. Once you get into the city, things get really complicated. Is that a plastic bag or a dog? Is that person going to cross the street or not? Those are the type of decisions that challenge the automated vehicles. The idea is “safety-first” (although Tesla has had plenty of problems). This means that if you are faced with something you don’t understand you slow down or stop.

      So not only would an automated streetcar be no safer than a bus, it would tend to be slower. A bus might needlessly slow down and change lanes to avoid that plastic bag. A streetcar would just stop and wait for it to blow away (which could take a while).

      1. A train is like a car on a freeway. It’s a fixed path, and you still a vehicle smart enough to stop when there is an obstacle in the path. It’s not like a vehicle on BlueCruise thinks “since dogs aren’t allowed on freeways, I’ll just ignore that dog sized object in my path.”

        “Is that person going to cross the track” is not a judgement a train operator (human or computer) needs to make. It assumes the path is free until it is not (as you said, if faced with something you don’t understand you slow down or stop).

        Sure, the train might come to a stop unnecessarily (like putting a traffic cone in front of a delivery robot), but that is mitigated by remote operations: there will be drivers stationed at the OMFs that can take over control and manually operate through unusual conditions. During regular operations, there will be qualified drivers constantly overseeing trains, watching video feeds (inside and out), etc., but the massive cost savings are allowing for 1 operating to oversee multiple trains, and the operators can take regular breaks because it is a desk job. (Perhaps during bad visibility, like a snow storm, Link would revert to manual operations with drivers in the cab).

        The important thing is dedicated ROW, which Link has. A streetcar in mixed traffic is as complicated as a bus in mixed traffic; the streetcar is slower than a bus, but that’s true automated or not.

      2. The important thing is dedicated ROW, which Link has.

        Yes and no. Unlike, say, SkyTrain (which is automated) there are crossings. It is these crossings that are complicated. Maybe not as complicated as what a bus has to deal with, but similar. The only major difference is how often you are likely to encounter a problem situation. Link is mostly grade-separated. None of our bus routes are.

      3. “Once you get into the city, things get really complicated. Is that a plastic bag or a dog? Is that person going to cross the street or not? Those are the type of decisions that challenge the automated vehicles.”

        Reality does not support this line of thinking. Waymo has 800 AVs operating in San Francisco every day (not all in the city of, but including some adjacent communities) – about half the size of the Metro bus fleet. In San Francisco they encounter situations much more difficult than any Seattle streetcar or what Link would face on MLK. SF is denser, has more pedestrians and bicyclists, narrower streets, busier traffic, etc. The idea that an AV will just stop and wait for a plastic bag to blow away is foolish. If it were at all true then every Waymo would be immobilized every day from trash on SF streets, or UberEats drivers on scooters darting in and out of traffic, or construction, or . . . . But the thing is the Waymo’s are not immobilized. They are encountering these problems and continuing to drive everyday. I think everyone commenting here is dramatically underestimating the incredible leaps in machine learning that have taken place in the AV world. “Is that a plastic bag or a dog” is a SOLVED problem for AVs and machine vision. Waymo’s are not just deciding if they have to stop for something in their path, they are evaluating oncoming traffic to decide when to make left turns, pulling to the curb and then pulling out again into traffic, waiting at crosswalks for people to get to the curb, etc. Automating the Streetcar or Link would not be a challenge for current AV technology, all you have to automate is throttle vs brake, when existing vehicles are automating not only that but steering, using blinkers, etc.

      4. Onux is right, this is a solved problem. Fully grade separated operations were solved decades ago, which is why old systems like BART and SkyTrain can drive themselves. When Link was initially built in the early 2000s this wasn’t possible, but has been solved and demonstrated in the real word in the past few years.

    3. Train automation also allows for a re-think of frequency. Right now Link drops frequency off peak to save money. With automation, running a 4-car train every 10 minutes is the exact same expense as running a 2-car train every 5 minutes.

      If Link is automated, rather than reducing TPH off-peak, ST can reduce VPH off-peak. Imagine late at night, 1-car trains every 4 minutes.

      Also much cheaper to dead-head vehicles, which can allow for longer span of service, as stations can shut down while empty vehicles run themselves to OMF.

      Link will probably still close overnight for track maintenance purposes, but much cheaper to provide overnight service, say run 24h hours on Fridays & Saturdays or a major holiday like New Years.

      1. Train automation also allows for a re-think of frequency.

        Yep. That is why various agencies around the world are building automated metros. It is pretty much the default now. It is why any new line in Seattle should be automated. For that matter, East Link should have been automated. We would probably need the same size trains (since they complement the trains from the south) which means the same size platforms. But if not for that little section in Bellevue it would be completely grade-separated and thus easy to automate. It gets a little messy because we can’t run trains down Rainier Valley very often but running the trains on both lines every 8 minutes all-day long would be doable and be cheaper for the agency than what they are doing now (or rather, what they are about to do).

        Oh, and automated train cars have more capacity. You don’t have space for the driver cab.

      2. Ross, that would only bring in a few new seats. But it would be nice to sit at the front and see the views similar to how you can sit at the front and see the views on a Community Transit double decker coach. Another good thing is that it would be more convenient, and faster. King County Metro drivers also don’t have to operate it anymore.

    4. Train automation is certainly possible and the steering logic is much simpler.

      Many automated systems have platform screen doors (like in SeaTac terminals). It’s not required but it is nice to have for safety.

      And of course, a train could fully run from Tacoma to Everett and there would be no need for driver breaks.

      I believe that an automated Link is inevitable. Link needs an automation plan for Link done before any ground is broken on any more extensions . It’s really antiquated thinking that they are still drawing and building new Link extensions without assuming that automation will soon happen. It’s as archaic as assuming new elevators will all have attendents back in the 1970’s would have been.

      1. “It’s as archaic as assuming new elevators will all have attendents back in the 1970’s would have been.”

        You’d have to take that back to the 1910s. Elevator attendants disappeared… I don’t know when, they were long gone before I was born in the late 60s.

        The Space Needle and such may be the last ones with an attendant, where they act as access control (can’t go to the top without a ticket, but can go to the restaurant), tour guide, and crowd control.

      2. I’m a decade older than you. As a very young child, I remember that many elevators had attendants. Wikipedia says that the change was mainly between 1950 and 1970. They rapidly started disappearing through the 1960’s and by the time I got to college in the late 1970’s attendants were very rare. I think you were perhaps a little too young to have lived through the transition but you barely missed it.

      3. I lived in the Manchester Apartments on Summit for a short time. The elevator is old enough to have a manual cage door but it’s still automated with floor buttons instead of an up/down/stop lever. I don’t know if it’s the original elevator but it’s very old.

        When I was in Russia in the 90s, many of the apartment elevators were the same kind: cage doors you opened manually, but fully automated with floor buttons, and they ran smoothly and reliably.

  4. Automated buses would be the biggest transit revolution since they invented trains. It has little to do with microtransit (or on-demand service). The dynamics don’t change that much. You still want an effective and efficient network. Some areas are cheaper to serve than others. But it would be as if we suddenly quadrupled service*. Think about that for a second. Every bus running four times as often. The great news is, it is actually better than that. There is an economy of scale that comes from transit (in contrast, it is the opposite with taxi-cabs or cars in general). Right now — even with a major funding shortage — we have buses running every ten minutes. Running four times as often is overkill. Thus some buses (like the 7 or a the RapidRide routes) would run every six minutes all day long. This means that other buses would get that extra service (and potentially run five or six times as often). You could easily see buses that currently run every hour transition to running every ten minutes. Of course some of the money would go into increasing coverage. Areas that require a really long walk to a bus would now have transit.

    But it gets better. Consider the trade-offs with two common transit patterns. With a grid pattern:

    1) Riders can get anywhere by making one transfer (while always going in the right direction).
    2) It is very efficient.
    3) Unfortunately there are frequent transfers.

    With a hub-and-spoke system:

    1) Riders can reach a major destination (the hub) without transferring.
    2) Transfers that do occur often involve detours.
    3) You waste service hours overlapping — or worse yet — almost overlapping. As a result, the buses run less often. This is bad for all trips, but especially those that involve transfers.

    This is why hub-and-spoke systems are more common when agencies have very little money. They allow transfers but they don’t expect very many. But the opposite is true as well. By running the buses more often, you minimize the pain involved with a transfer. You reach a “tipping point” where a grid is a much better option. Riders would be expected to transfer — even for common trips — but they could get anywhere to anywhere without time consuming waits or detours. Thus if they quadrupled service, riders could see a much higher effective increase. You would have a much better network with more frequent service and better coverage. Of course it wouldn’t happen all at once but it would be a huge improvement.

    *I’m basing that number on the idea that driving the buses accounts for 75% of the operating cost of a typical transit agency.

      2. The purpose of drone vehicles is for someone to make money. Everything else is incidental, at best a lucky side effect.

      3. The purpose of drone vehicles is for someone to make money.

        As opposed to every other private company in the world — including the companies that make buses and trains. My point is, so what? So what if a company decides to make money off of automated buses. If it improves transit then it is a good thing. Again, this is far more likely to happen first in Europe (or Asia) then it is in America. You can’t expect us to be a leader in any aspect of transit. But as Europe starts running more automated buses, so should we.

      4. ‘The purpose of drone vehicles is for someone to make money.’

        You won’t support any public improvement because somebody will make money off it? The issue is not whether they would make money; it’s whether the public benefit is real and substantial. Ross just said how auto buses might allow us to go from mediocre frequency to excellent frequency. That’s something we need to do anyway.

    2. Oh, and it could also lead to a change in the way that the federal government helps local transit agencies. Historically, the feds have chipped in for capital projects while the local agency paid for operations*. Unfortunately, this encourages dubious capital projects along with poor service. You can get a grant to run a streetcar carrying a few hundred riders a day while the agency doesn’t bother to run their buses very often. This in turn has led to mode fetish by the various agencies as the feds favored one technology over the other. Light rail (instead of heavy); trams; Bus Rapid Transit (instead of just running the buses more often). Each mode has their place but the country seems to following fads more than making smart decisions. This helps explain why we have such bad transit in this country. We are ignoring the basics (like good headways).

      But automating the driving could change all that. Once you automate the driving, the focus is on maintaining the buses. But what if you want to run the buses more often? You need to buy more buses. This is where the federal government could chip in — they could help pay for more buses. There would still be major projects (like railways or busway) but the agency would have to show that it would be better than just buying (and running) more buses.

      Of course that could happen right now but it wouldn’t make that much difference. Most agencies aren’t spending a lot of money on buses (or even maintaining the buses). They are spending it on driving the buses. Right now, if an agency got a grant to buy a bunch of new buses they could end up gather dust most of the day. In contrast, if the driving was automated they would definitely be used a lot.

      *I’m not sure why the government grants are based on capital spending versus service. My guess is that they were afraid agencies would just shift service. Think of a city-run transit agency. They have other needs (police, fire, social services, etc.). If they get a grant to run the buses more often they might do just that. But they could also just use the federal money to retain the same level of service while spending more money on the police, fire, etc.

    3. A mix of hub/spoke and grid would be ideal. Our geography makes grid a little complicated at times.

      Hub/spoke works better for suburbs, while grid works better for the city. On top of the network, having freeway coverage will help people get across longer distances between hubs. Local service can merge onto the freeway to provide parallel service to the corridor along freeway stations and major hubs.

      Automation is great since it can time the ride exactly as scheduled. There should be little margin in arrival time. That means automated buses should be able to move with no impediment other than stopping / dwelling times.

      1. Our geography makes grid a little complicated at times.

        Absolutely. I’m speaking in generalities here. If you run all the buses more often, then transfers aren’t as bad. If transfers aren’t as bad you can make your system more grid like. This is another way in which an increase in ridership (say doubling the frequency) can result in more enhancements than you would assume.

      2. An example is when the all-day 43 was eliminated and the migration path was transferring from the 48 to the 8, former 11, or 2. But the 8 stop is 3 blocks away around the corner. The 11 was a block down a steep hill. The 48 was 10 minutes but when Seattle’s TBD was reduced it went down to 15 minutes. The 8 is 15 minutes daytime, 20 minutes evening. The 2 and 11 are 30 minutes evening. The 2 was hourly evening for a while. So none of the transfers was great, you had to look up the schedules and choose which route to transfer to ahead of time, and the bus stop was a way away. So these transfers aren’t very practical for everyday use. If the routes were more frequent, that would help a lot. Now RapidRide G is frequent so that’s something. But it goes to Madison, further south than the others.

  5. There are issues, and you can’t expect an American agency to be a leader. Things will happen in Europe long before they happen here. This includes how they deal with the unions. My guess is there will be a mix of phasing out driving and paying off drivers. Some will transition to be “fare ambassadors”.

    As the technology matures, it would make sense to apply it first with our RapidRide lines. These already have a proof-of-payment system. They also account for a high proportion of our driver hours (because they tend to be frequent). Ultimately it would make sense to go with complete off-board payment, even during the transition (you could have complete off-board payment even with a lot of drivers).

  6. Unfortunately, what can actually be done to modify existing equipment is limited not by what’s physically possible, but by what one company (e.g. the manufacturer) chooses to support. And companies usually prefer to sell new stuff, rather than upgrades to existing stuff.

    In practice, my assumption is that automating a train would require buying whole new train cars, and automating busses would require buying a whole new bus fleet. And, of course, for buses, there would be weird volume-based restrictions like not being able to combine automation with left-side doors, trolley poles, or any other niche features.

    While it may theoretically be possible to bolt automation onto an existing bus fleet, I suspect it’s not possible in practice, other than, perhaps, a humanoid robot sitting in the driver seat, physically controlling the pedals and steering wheel (which would not be nearly as safe or reliable as existing automated vehicle systems).

    2. automating busses would require buying a whole new bus fleet

      Yes, but that occurs anyway. Most agencies replace their buses every dozen years or so. Again, this wouldn’t happen all at once. It is far more likely that you would have some routes being automated while the rest stay the same. This makes sense from both a labor and equipment standpoint. But over time you would see the transition.

  7. I see automated buses rolling out first at slow speeds. To me, the issue is more how fast they can go, rather than if they can operate at all.

    I suspect that they’ll first be used and probably most practical to use as a replacement for Metro Flex service.

    I see the next big challenge he will be with rider safety in the vehicles. Should standees be allowed? Is there an ADA violation if they’re is no driver/ attendant to secure a wheelchair? Will riders avoid them out of security fears, especially if there are some high profile crime incidents?

    There are at least a dozen or two entities building and marketing the technology — so we see that it’s technologically possible and potentially profitable for a manufacturer. But it’s the implementation in an urban context that I think will be where challenges are.

    1. I see automated buses rolling out first at slow speeds.

      There are actually two environments where automated vehicles are relatively advanced and safe. They are on the opposite sides of the spectrum. It is common to now to have “hands free” operation on the freeway. But the vehicles are still fairly dangerous on the city streets. Like all driving, the slower the vehicle, the safer it is.

      Then you have to factor in the labor savings. There is no point in automating a bus like the 225. It is so infrequent and short that the savings would be minimal.

      This combination of factors could lead to some interesting route choices. I could easily long-distance express buses being automated first. Think of the 594. It spends most of its time on the freeway. It spends a lot of its time in Downtown Tacoma or Seattle (where speed limits are low). Thus even if it went slower than a driver would in other parts of town it would still be fairly good service. Yet the savings would be enormous (even if you increased frequency).

      1. “But the vehicles are still fairly dangerous on the city streets.”

        This is just not true. Waymo, Zoox, others are all operating on city streets in very challenging environments like San Francisco with a safety record much better than human driven cars. The case in the news right now is where a Waymo hit a child that walked into the middle of the street from behind a parked car. The Waymo braked immediately and went from 17mph to 6mph. The child was knocked over but got up and walked right back to the curb. The AV sent an alert so the company could call 911, then waited for the police to arrive. A human driver would probably have driven right over the kid without enough time to react, to say nothing of hit and run drivers or people speeding faster than 17mph in a school zone to begin with.

  8. Trial these on semi-closed environments first, such as between resorts at Disney World, terminals at airports, or a spread-out college campus. More allowance for error and chances to fine-tune the technology, with a reduced chance of property damage and human injury, than on a city street.

    1. Correct. There are some current trials going on at various college campuses. That’s a well contained and fairly safe environment.

      1. You are well behind the times, Waymo has been operating on city streets in San Francisco for two and a half years with customers. They are now alo in LA, Phoenix, and Austin. Fire more major cities are planned for this upcoming year. We are well past the “trials in contained semi-close environments stage.”

      2. Onux, I’m talking about transit, not taxi operation. Autonomous taxi operation has made a lot of progress, the sensor technology is the same, but operations is slightly different if you want to allow many riders to share a bus along a fixed line.

      3. I don’t think the transit riding public or news media would tolerate all the safety problems that Waymo continues to have. Yes, there’s been progress, but having a bus collide with something and then block a major street like Waymo blocks this minor street would probably get the technology banned from the transit industry for decades:
        https://m.youtube.com/watch?v=PvLU8lhkx6I

        The transit industry can’t afford to be a playground for the tech industry.

  9. Would there be value in having automated bus routes designated with paint or a series of reflective little pavement domes. They would visually alert pedestrians to the presence of an automated bus as well as possibly provide guidance for the bus.

    It would also seem to reduce the amount of “ learning” and processing that a vehicle would need as they would be following a fixed route.

    Along with that, I would think that it could be useful for the vehicles to have a unique way that they’re illuminated in the dark. The guidance on the roof makes it obvious during the day, but that’s not easily observable at night. For example, alternating brighter headlights (think electronic “wigwam”) would give pedestrians advance notice. Another example would be an LED strip that changes color when doors open like the newer Link trains do.

    1. My understanding is that if not for the other cars and people, the vehicles can drive themselves just fine. It is very counterintuitive. We assume the hard part is steering and staying between the lines (like someone learning how to drive). For the machines, that isn’t the hard part. They can parallel park like a pro. The hard part is traffic an differentiating between objects or dealing with the unpredictability of vehicles or pedestrians.

      I could see value in notifying people that an automated bus could be coming that way. But any vehicle is inherently dangerous and we are seeing an increasing number of “regular” vehicles being operated with little or no control by the drivers.

    2. This might have been a valid strategy a decade ago, but technology has leapt past the point of needing route marking assistance. Waymo and Zoox are operating fully autonomously on all kids of city streets, with patched pavement or metal road plates, centerlines or no center lines, taking a different route for each trip and dynamically re-routing around obstacles or road closures, etc. They don’t need special paint and there is more than enough processing power to handle everything.

      1. Yes of course route indicators on the street are technically no longer needed. However, they do offer two good things:

        1. They could more easily achieve exact curbside alignment (“no gap”). That is a boon not only for wheelchair accessibility, it enables easier loading and unloading.

        2. It would give an extra level of visible awareness for the vehicle. It would more prominently indicate “there’s a bus running right here” than a mere bus stop sign or even an enhanced stop with shelter and garbage cans does. In other words, it’s like having a continuous billboard for transit. We technically don’t have to paint exclusive bus lanes red either but it sure makes a statement.

      2. Onux, there is a big difference whether you’re running a taxi service for a dedicated trip or you run a fixed line service along a specific route. Route based services allow for more riders per bus and you don’t need an app, you can just wait for the next bus to show up.

  10. Autonomous buses are even easier than autonomous cars.

    They follow a fixed route that they can be effectively trained on. The roads can be engineered to support their operations and provide a clear, definite track.

    With TSP and red paint, autonomous buses are quite easily to implement. The only “hard” thing is dealing with other drivers, pedestrians, and unexpected elements on the road – which we have technology for. We could have an operator on board to take control when necessary until confidence builds.

    It takes willpower to make things work.

    1. But yes, we should prioritize train automation. It’s far easier and very doable on the 1 and 2 Lines.

  11. I have pondered if the automated bus technology should be first used for last-mile access, with major routes having stop spacing more like light rail. They could be positioned to be idle at “stations”, where a rider could have reserved a seat beforehand with an app and would be directed to a bus upon arriving at the station. Similarly, the vehicle could be hailed from an app and the bus would take that rider to a nearby designated “station”.

    That’s seemingly impractical for busy Seattle bus routes, but it could make transit more accessible across the suburbs. For example, a Star Lake hub based in the station’s garage could offer riders a choice to hail an automated bus to reach places in the catchment areas of Kent and Federal Way. It’s got to be cheaper than adding a new 1000 space garage — and better for the environment and for TOD opportunities .

    1. Part of the problem is not every bus trip out in suburbs is connecting to Link. That’s one use, but there also people that just want to make local trips between places that Link doesn’t go. So, it’s really a classic question of network tradeoffs, not automation.

      For example, to provide the best possible Link connections, you want a hub a spoke network centered at the Link stations. But, for just making local trips that don’t involve Link, such a network is terrible, as trips that are short and straight in a car require detours to Link stations and/or transfers.

      To run both networks simultaneously, you have to cut frequency to pay for it, frequency which is already probably very poor, to begin with. Hence, tradeoffs.

    2. If you can have it for “microtransit” type service then you can have it for a fixed route, fixed time bus. The latter is bound to carry way more riders and save the agency way more money (if it is automated). Part of the reason agencies have invested in microtransit is that is transit theater. They can claim to serve a big area but in reality, the wait time is terrible and a relatively small number of people actually benefit. In contrast, consider what Metro would do if just the RapidRide E was automated. That is a huge amount of service hours that could be put into running other buses more often.

      To be clear, it would be a good idea for an agency to invest in bus service before building a giant garage. But that can be done via regular routes running more often.

    3. These can be combined. The taxis have to wait somewhere for a call. If they wait at a Link station, they can serve both Kink and non-Linj trips. Metro Flex has three cars waiting just south of Rainier Beach station for the southeast Seattle area.

  12. And what if that bus was a 10-15 passenger, cheap, and far more frequent. No tearing up roads, better service, during busy times run more buses.

  13. Southboubd train came. Somebody said “Yaay” so they may have been waiting a while. Anybody know what else is going on?

  15. I’m home now and looking through ST’s alerts. It was a “non-involved collision” between Othello and Rainer Beach stations. The alert says it was resolved at 6:04pm but the last station announcement at 6:30pm said it was ongoing. I assume it’s back to normal now.

  16. re: Link automation

    From my understanding, Northgate to Judkins Park is all grade separated and has pocket track on each end.

    Is it possible to run an automated Link overlay between Northgate and Judkins Park with fast frequency?

    I’m reminded of that Link station ridership map that’s almost like a fat snake between the stations: https://www.theurbanist.org/wp-content/uploads/2023/11/Projected-Link-Ridership-edited.png

    This could be a good way to incrementally automate the system

    1. Is it possible to run an automated Link overlay between Northgate and Judkins Park with fast frequency?

      I don’t think so. You would still be running trains from Redmond to Lynnwood along with trains from Federal Way to Lynnwood. Basically those other trains will combine for five minute headways from downtown to Lynnwood. The overlay would have to run every five minutes, for a combined 2.5 minute headway along the same section. This is problematic, based on this: https://seattletransitblog.com/2015/03/21/capacity-limitations-of-link/. To quote the article:

      Although going below 3 minutes is possible, due to the variability inherent with human factors and surface operations it “wouldn’t give our ridership as reliable a service.” The small windows to fit in delayed trains might cause them to bunch up, delaying riders. Furthermore, it would likely require additional investment in Traction Power Substations.

      The one time when it could work (without a lot of effort) is at night. Southbound, the trains switch over to running every 12 minutes starting a little after 9:00 pm. So you could run an overlay and provide 3 minute headways through the core (and 6 minute service to Judkins Park). After 10:00 pm the trains start running every 16 minutes. This would be easier and more valuable. The train would provide 4 minute headways (and 8 minutes to Judkins). I’m not sure if it is worth the effort.

      1. I see, I didn’t realize Sound Transit was looking at a 3 minute headway floor. I’d heard that 90 seconds was the theoretical minimum. And requiring more power stations makes it a no-go for me too

      2. Actual and scheduled headways are different things. A scheduled headway of 3 minutes means that sometimes trains will come less than 2 minutes apart. Not only is there the effect of traffic signals in spots, there’s also the time uncertainty involved if train doors need to be held open at several stations due to things like crowding in front of the doors.

        So a 2 minute scheduled headway won’t have any reasonable time cushion. It will create a backup even with minor delay to just one or two trains like a traffic signal on MLK.

        The need to offer anything below 8-10 minutes to me is based entirely on crowding rather than pure waiting inconvenience. I would even point out that transfers between lines takes time, especially at planned ST stations in DSTT2 to negotiate the deep Chutes and Ladders station layout will make every transferring rider take more time in transferring than waiting 4 minutes more (2 minutes when averaged) at the initial boarding.

      3. “I didn’t realize Sound Transit was looking at a 3 minute headway floor. I’d heard that 90 seconds was the theoretical minimum.”

        3 minutes is ST’s policy limit in the DSTT currently for regular service. Beyond that it’s worried about train bunching, signals, platform crowding, and needing more egress escalators/elevators/stairs. It does run 1.5 minute trains after ballgames between Stadium and Roosevelt, but the schedule goes out the window then.

        The tunnel can be upgraded for 1.5 minute regular service. There was a candidate project for it in the run-up to ST3, but it was deselected when DSTT2 was selected. We’re trying to get ST to revive it for a no-DSTT2 alternative in ST3.

  17. I’ve pondered whether or not it’s viable to have a bus platoon with a driver in a front bus, and an automated bus or two closely trailing that bus (and controlled by the front bus driver) to create a “train” so to speak. It’s a different way to introduce automated buses and it seems like it would be much easier to implement.

    Imagine every RapidRide bus having a second driverless bus right behind it, responding to the driver in the front bus as if they are both attached as a train.

    1. I think that is how they run buses in Zion National Park although they are attached via a wire. Same idea. Basically just two buses acting like one giant bus. Of course we could always just get a bigger bus. It only makes sense to do this sort of thing if you have a huge load to carry. We really don’t have that anymore. None of our buses are that crowded. They peak out at maybe five minutes and that is only for a brief period.

      There are other strategies for dealing with the crowding. One is an express overlay. The 15 is a great example. It skips several stops and skips Uptown. It basically poaches some of the D Line’s riders. But if the D Line is running every couple minutes, that doesn’t matter. It is better for both the agency and riders. Running the 15 is cheaper than running another D. Some riders prefer the express while you retain good frequency on the core route. Unfortunately we aren’t at that level of crowding. The D Line runs every 7 minutes. We could run the 15 a few times but if the money came from running the D Line less often it wouldn’t be worth it. This is a long way of saying that capacity isn’t a big concern with our system right now.

    3. Munich plans to test platooning buses. You get more flexibility during rush hours while a single bus I sufficient at other times.

  18. Bus drivers do a lot more than just drive the bus; fare collection, answer questions, secure wheel chairs, act as some sort of rule enforcement (e.g. no smoking). Link for sure could be automated but not sure the cost savings would justify the capital expense. Not to mention union battles. I could see ST Express buses being automated on long freeway runs. Especially the ones where half of the driver hours are deadheading the bus to/from the terminus. Lets wait until the trucking industry automates semis before talking about buses.

    The services like Access should probably just contract with the robo taxi companies. But there’s a whole industry where transit dollars get funneled to not for profits like Hopelink that would have to be overcome.

Comments are closed.