Progress and Public Transit, Part Four: Vehicle Automation

That which can be automated, should be

Transit has been caught in an Endless Emergency of rising costs for inadequate service for decades.

The first essay in this series explained why this is so. To briefly recap: failure to recover costs has locked transit into permanent reliance on external subsidy. That arrangement creates warped incentives throughout the system. Without need to keep costs below fares, politicians can insist on unsustainable routes and coverage; labour faces no check on its demands for increased wages and benefits; and good managers are handcuffed while bad ones always have tools to obscure their poor performance.

Worse, neither decreasing nor increasing subsidy can change this dynamic. More funding causes brief respite, while less funding makes things worse, but the system always reverts to the mean. The only way out is cost recovery: a transit system must make more money in fares than it costs to operate. Only under such conditions do everyone’s incentives align properly.

Today, in the last instalment in this series (for now), we will explore how technology offers a way out of the Endless Emergency. Technology might help transit lower its operating costs such that farebox revenue could cover them.

You may wonder why we didn't start this series here. Changing Lanes is a techno-optimist newsletter that sees technology and innovation as the tools that will help us build a better world. Isn’t applying tech to any problem our first resort?

No.

It’s this sort of thinking that gets you Hyperloop: applying solutions before you understand the problems is typically a waste of time and money. It’s such a common, and seductive, mistake, that it has its own name: the gadgetbahn temptation.

To avoid falling into it ourselves, the middle entries of this series considered how we could solve the Endless Emergency with the tools we already have. We know how to reorganize institutions to align incentives better, and we should do so. We know how to align our transit networks with dense urban form to ensure they can support each other, and we should do so. As is often the case, it’s not that we don’t know what would work, it’s that doing what would work is difficult.

But for today, we will leave the policy wonks to continue the struggle for institutional reform, and the YIMBYs to push for real transit-oriented development, and turn our attention to what the technologists have to offer: vehicle automation. Just as in every other field, we can substitute the human labour that operates transit vehicles for machine labour, and reduce the cost of the service even while operating more efficiently, precisely, consistently, and safely.

We know this, because in some situations, it’s already happened.

Rail Automation Now

In the 1970s, the government of Ontario believed that automation was the future of transit.

It founded a government-owned firm (in local parlance, a Crown corporation) to design and develop new rapid-transit technology for areas dense enough to need it, but not so dense as to require a full subway. This new tech was called the Intermediate Capacity Transit System (ICTS): light-rail cars operating on an elevated guideway. And because it had its own right-of-way, and only needed to go forward or backward, it was indeed capable of fully-automated operation, with no driver necessary. The developers were ultimately only able to sell ICTS to three markets: Detroit (People Mover), Vancouver (SkyTrain), and Toronto (Scarborough RT).

Vancouver began SkyTrain service in 1985 and still operates it. In all that time, it has never had a driver, and there has never been a fatality.

Detroit began using its People Mover in 1987 and still does. In all that time, it has never had a driver, and there has been only one fatality.¹

Toronto began using the Scarborough RT in 1985 and closed it permanently in 2023. In almost forty years of operation, it always had a driver.

A rough Fermi calculation says that the Scarborough RT, during its lifetime, had almost 4 million trips where a driver accompanied the train. Or put another way, the TTC retained the services of perhaps 27 drivers, every day, for almost 40 years.

As comparison with Vancouver and Detroit shows, the drivers were unnecessary and their contributions superfluous.

The contrast between systems is instructive. Rail-based transit represents the easiest, fastest win in transportation automation. Operating in controlled environments with predictable movements, automated rail systems can do everything a human operator can. Indeed they can do more: they can dock at stations precisely, reducing platform gaps, and permitting station doors to be installed, reducing the chance of injury or death.² They can safely run shorter headways than a human can, permitting more-frequent service. And they can save on operating costs (those 27 drivers each day weren’t working for free).

Around the world, there are fully-automated subways (Europe and Asia) and freight rail systems (Australia). Passenger heavy rail is in development. Those who want to see progress in public transit should advocate for automating rail systems as the first priority: new systems should be automated from the beginning, and old ones retrofitted. That retrofitting is, relative to the benefits it offers, remarkably inexpensive. The principal requirement is a modern electronic signal system, which has an order-of-magnitude cost of the hundreds of millions of dollars; in other words, the amount it costs to build a single kilometre of a new North American subway.³

Automated Buses Next

Rail automation is the first priority because it is easy. The second priority is bus automation, because that’s where the money is.

It may surprise you to learn how expensive bus drivers are. In 2017, by one reckoning, the driver’s wages represented 42% of the average cost of running a bus for an hour. Transit expert Jarrett Walker estimates the figure even higher, counting the driver as 70% or more of the average cost to operate a bus.

This means that automating buses would offer substantial savings. To look at a specific case: labour makes up 65% of San Francisco BART’s fiscal-year 2025 budgeted operating costs ($726M out of $1,116M). If the buses were all automated, and we could apply a 42% savings to this figure, BART would save $304M in 2025. As it happens, that figure is close to the $342M BART is requesting in emergency subsidy from the U.S. federal government to keep operating next year. And those savings would apply in all future years as well.

It couldn’t do it on its own, but combined with other changes, automating the buses could help BART achieve full cost recovery, halting the Endless Emergency.

There will be questions. The first is, can this be done—is it possible? The second is, should this be done—might it cause more problems than it solves?

Yes, it certainly can be done. Different firms and governments have been experimenting with automated buses for decades. After all, buses generally follow one route, rarely or never deviating from it, and such predictable situations are exactly what machines excel at. All that is necessary is some means for the bus to know where it is in space and where it is going next.

In the past, engineers have tried to make this work in a few different ways. Some buses were fitted with cameras and followed a track of special paint on the road. Unfortunately these tracks could be covered in snow, or erased by wear-and-tear. Other buses were fitted with magnetic sensors and followed a wire embedded in the pavement; this is the approach favoured by the Chinese ‘trackless tram’. This approach isn’t affected by the weather but has proved persistently glitchy.

But that was the past. In the present, we have Waymo.

Waymo’s success in deploying robotaxis to city streets, in all their messiness, shows that operating an automated vehicle in mixed traffic is possible. Less well-known but possibly more relevant is Gatik, which offers middle-mile goods movement via automated truck. Gatik maps its routes in advance and proceeds slowly and carefully along pre-planned corridors, to minimize novelty; exactly what we would expect a bus operator to do.

To the best of my knowledge, Waymo isn’t aspiring to build an automated bus for transit operators. Neither is Zoox nor Cruise. Tesla is building the Robovan, but given the CEO’s notorious disdain for public transit, it’s unlikely the firm aspires to supply rolling stock to traditional operators.

The firms’ intentions today don’t matter, though. What does matter is that the same systems—software and hardware—that power contemporary robotaxis and self-driving delivery trucks could also automate a city bus. Indeed, that hardware could be supplemented with the retro optical and magnetic automation tools, to ensure redundancy. Better yet, put the bus in its own right-of-way, as Cavnue proposes, and you have a fully-dressed system for bus automation.

So it can be done. Should it?

Let’s review that question after we consider the possibility of replacing buses with robotaxis outright.

Robotaxis on Feeder Routes Soon

There are two ways we might do this.

In the weak version, robotaxis would replace only the bus routes with the lowest ridership. These are typically ‘feeder routes’, which connect low-density neighborhoods to major transit corridors or hubs. Such routes provide coverage, which are often part of an operator’s mandate, but are expensive to operate. They run nearly empty most of the time, yet still require a full-size bus and driver. Instead, the transit agency could eliminate the buses, and provide on-demand shared rides in small-capacity shuttle vehicles.

Because such an approach doesn’t require a robotaxi—all it needs is a vehicle, a driver, and a way to call for a pick-up—many agencies have been experimenting with this approach for a while. If you use it to replace entire feeder routes, it’s called ‘microtransit’; if you do it but only for the benefit of passengers with mobility needs, who face barriers to using conventional bus service, it’s called ‘paratransit’. A number of firms, including Via, Spare, and Argo, partner with traditional agencies to provide it. By using software to optimize routes and combine different passengers' trips, agencies can serve more people at lower cost than running near-empty buses on fixed routes. Automating these vans is an obvious next step.

But there's also a stronger version of the proposal: in some contexts, robotaxis could replace all transit service. Most American transit systems outside a handful of dense cities are, in practice, a social service for those who cannot drive. They run infrequent buses on circuitous routes, trading off ridership for coverage, serving few at high cost. The result is the Endless Emergency in a nutshell, described succinctly by Matt Yglesias: “ridership is extremely low, which means that the cost per rider is very high, which means that improving frequency or expanding service is prohibitively expensive, which means that the quality of service is terrible, which means ridership stays extremely low.”

Transit agencies could transform themselves into mobility managers rather than operators.

In such contexts, providing subsidized robotaxi service is extremely attractive, at least at first glance. The operating cost would be lower than running near-empty buses. The service would be dramatically better: door-to-door rather than requiring walks to stops, on-demand rather than hourly schedules. And the social service function could be preserved through targeted subsidies, as we’ve discussed before.

These targeted subsidies—provided not by the transit operator, to avoid misaligned incentives, but by social-service agencies or the government directly—could provide support to specific demographics, times, and locations, calibrated to achieve public policy goals, such as maximizing vehicle occupancy, reducing parking demand, or ensuring service to mobility-disadvantaged populations. In this way, subsidies could shape how private mobility providers serve public needs.

More boldly, transit agencies could transform themselves entirely, becoming mobility managers rather than service operators. Instead of running their own vehicles, they would oversee fleets of automated vehicles provided by private firms, using subsidies and incentives to ensure these fleets serve public purposes. The vehicles themselves might range from automated microtransit vans that aggregate multiple passengers headed in the same direction, to individual robotaxis for areas or times when demand is too sparse to justify shared service. The agency's role would be to ensure that this privately-operated service meets public needs: universal access, affordable fares, and comprehensive coverage.

For more on both ideas, you may consult the past work of John Niles and Bern Grush; or better yet, wait till next year and consult The End of Driving, second edition (forthcoming 2025 from Elsevier), co-written by Grush, Niles, and myself. Our book will make the case for the integration of robotaxis with public transit, at length.

In the meantime, remember that this isn't a proposal for New York or Chicago or Toronto, where the ridership and geometry requires mass transit. But for mid-sized cities and larger towns like Bangor or Bakersfield or Brandon? For the lowest-density suburbs? For rural areas? The robotaxi would be a better solution for most routes in these places. Traditional fixed-route transit could be reserved for corridors dense with origins and destinations, the sort of place where the service can recover its costs from fares.

Overcoming the Barriers

Vehicle automation is a promising tool for addressing the Endless Emergency, for a few reasons.

Firstly, introducing it would be simpler and easier than the other approaches discussed in this series. Transforming transit into a regulated utility rather than a public agency would involve picking fights with almost everyone: the unions, the non-union staff, the politicians, and lots of riders, all of whom would stand to lose something in the short term. And transforming our cities will require us to pick fights with homeowners, the most powerful constituency in municipal politics. Conversely, choosing one kind of vehicle, or one kind of route, and automating it, requires no institutional change and only involves starting two fights: one with the unions and one with the regulators. Winning those won’t be easy, but it would be easier than the alternatives.

Secondly, focusing on automation allows us to do what we’re good at. As Tyler Cowen pointed out recently, focusing on mobility rather than density plays to North America's strengths. Our cities are already optimized for travel by cars on streets, and for the most part are not optimized for density. Leveraging our existing autocentric infrastructure, using it better rather than trying to replace it, is the faster, more efficient play.

That said, there are concerns to keep in mind. Those are cost; labour relations; and safety.

For cost, we already have real-world evidence that should give us pause. While not involving automation directly, the experience of Innisfil, Ontario offers important lessons about the economics of replacing traditional transit with on-demand service. I’ll write up a full newsletter on this some day, but the short version is that the Town of Innisfil (a Toronto exurb) has skipped building a conventional transit system in favour of subsidizing Uber rides for residents. The theory was sound: rather than spending millions on fixed-route buses that would serve only a small portion of the low-density municipality, the Town would pay for direct trips that people wanted to take, and only those. And initially it worked well. Within eight months of launching in 2017, over 3,400 residents had taken more than 26,700 trips. In fact, it worked too well: latent demand for trips was far higher than expected, straining the budget. To limit demand, the Town restricted trips to those that begin or end at key civic locations. Even that has not been enough, and Innisfil is now exploring adding fixed-route transit to complement the on-demand service.

We can afford to pay off the people who will lose out, but can't afford to abandon economic growth from automation.

The Innisfil experience is encouraging, because it shows how, even in low-density places, there is significant untapped demand for better mobility options. But it’s also discouraging, because it shows how expensive fulfilling that demand is, even with a digital layer and an on-demand model. For robotaxis to completely replace transit, the per-trip costs need to be dramatically lower than what human drivers can provide; as Jarrett Walker notes, the cost of delivering it is the principal constraint on microtransit today. Just how cheap robotaxis can get at scale is not clear yet, which means whether they can get cheap enough is uncertain.

Labour relations are another problem. Unions have every reason to oppose automation, given how many jobs it is poised to eliminate. Unions exist to protect their members, and so naturally they will resist as hard and as long as possible.⁴ That said, the purpose of public transit is to improve mobility options, not to give jobs to transit employees. The fact that the Scarborough RT never operated without a driver is deplorable; and we’ve seen similar outcomes at North American ports and shipbuilders for decades.

In all cases, I think the solution is the same: a buyout. Promise no layoffs; retraining; workforce reductions by attrition; and, where necessary, straight-up buyouts of existing workers. We can afford to pay off the people who will lose out, but we can’t afford to abandon the economic growth that automation offers.

Finally, there’s safety. Bus drivers do more than drive. Depending on the system, they may also collect fares, respond to emergencies, and maintain order on the vehicle. Indeed their very presence may serve as a security measure; the absence of an authority figure may encourage vandalism, harassment, or other vexatious behaviour. If we eliminate drivers, does that mean these needs go unmet?

Certainly not. As noted earlier, some transit lines have been wholly automated for a while. Vancouver's SkyTrain and Detroit's People Mover employ attendants at stations, who can respond quickly to issues on vehicles. These are supplemented with roving attendants; remote monitoring by cameras; and onboard communication systems allowing passengers direct engagement with the operator. Further, as I’ve noted before, when robotaxis become mainstream, I fully expect some of them to be vehicles that contain private pods, so the trip, but not the space, is shared. Multiple layers of safety oversight are possible while still capturing the benefits of automation.

The Emergency Can End

This series has about progress and public transit. What should believers in progress expect to see transit become, and how can they help that change to happen?

I have argued that progress means halting the Endless Emergency. The first essay explored this dynamic in detail, showing how operators’ dependence on subsidy creates perpetual crisis. The second essay considered reorganizing transit agencies as regulated utilities, allowing them to charge what the service truly costs, would help to break this trap. The third essay examined how we could also reshape our cities to provide density at transit stations: a solution which would not only help to make transit financially viable, but would also make our cities more liveable.

Today’s entry has explored how vehicle automation offers another opportunity to fix the problem. By dramatically reducing operating costs, automation could help agencies achieve full cost recovery. Rail automation is proven and cost-effective but politically challenging, as the contrast between Vancouver and Toronto has shown. Bus automation holds enormous promise given how much of operating costs go to drivers, but the technology is still maturing. And while replacing low-ridership routes with automated on-demand service is appealing, Innisfil's experience with human-driven ride-hail shows that making the economics work may be difficult.

These challenges are real, but they're not insurmountable. It’s a complicated programme but we can make progress on all fronts simultaneously. Our advocacy should be for institutional reform of operators; zoning reform in cities; and technological reform wherever possible, beginning with automation of rail systems. By pursuing all of them, we can travel a path that leads to transit systems that can sustain themselves financially while providing the service our cities need.

For all these measures, the best time to begin working toward them was decades ago. But the second-best time is now.

---

Thanks for reading Changing Lanes! If you’d like to respond to this post, please leave a comment below.

And I hope you’ve enjoyed this series on Progress and Public Transit. The series is not over: future entries will consider other promising ways to stop the Endless Emergency with technology. One notable way is with what I call PUPPET systems: Personal, aUtomated, Point-to-Point, Elevated Transit. Expect a post about this in the New Year. But for the rest of this year, we will turn away from public transit to other topics.

1. In 2016 a patron fell between the cars while the train was stopped at the station, and was killed when the train automatically began to move.

2. Indeed this is what Detroit did in response to the fatality referenced in the previous footnote: the operator installed bollards on the platform to block access to the inter-car spaces.

3. In the last entry in this series I mentioned the Swiss proverb about rail improvements, electronics before concrete. It was coined for situations just like this one: it is better to install cutting-edge signals to use existing infrastructure more efficiently than to build new, expensive infrastructure, especially tunnels.

4. I don’t begrudge them this: at Changing Lanes, we hate the game, not the players.