I’m going to try something new for this blog (and something I haven’t done since at least my first year of graduate school) – a book review. I picked up Anthony Townsend’s Ghost Road: Beyond The Driverless Car shortly after it was published last month, and after blowing through it over the July 4th holiday weekend, decided to share my thoughts.
When I joined a research team at Georgia Tech contracted by the Georgia Department of Transportation to write its AV planning roadmap in 2017-2018, I spent hundreds of hours reading books on past innovations in transportation and telecommunications, white papers from consultancies, financial institutions, and commercial real estate firms, roadmaps already published by other state DOTs, and all the academic research I could get my hands on. Hypothetical AV development timelines. Deployment scenarios. Social and business impacts. Most of this literature is utterly forgettable, and surprisingly unimaginative. However, Townsend’s “Reprogramming Mobility” scenarios paper struck a chord with me in a way most AV papers haven’t. Even written in 2014-2015 (which now feels like ancient history), it remains more multi-dimensional than other illustrations of “our AV future.” It doesn’t bother with polar “heaven and hell” scenarios, but looks at various combinations of political and geographic settings, medium-term timelines, and key actors that might shape the evolution of transportation systems. And despite its age, it’s the first paper I still typically refer to friends who are interested in learning more about AVs. Therefore, I had high expectations for the book, and wasn’t disappointed.
Townsend’s strength is his ability to introduce these dynamics while rarely putting his own thumb on the scale. He veers into forecasting an industry segment’s endgame just long enough to illustrate a use case for automation, the c. 2019 state-of-the-art, and how it might look if widely adopted. Unlike those “heaven and hell” scenarios we often see urban planners project for alternative AV scenarios…where everything is either terrible because the private sector and entrenched institutional players are allowed to run amok, or amazing because Good Government steers urban form decidedly away from cars…Townsend offers critical analysis as to why certain deployment scenarios are likely or unlikely to play out. He also demonstrates a deeper understanding of the technology’s inherent limitations than most academics or planners, shaped by industry developments of the past half-decade.
He seldom sorts an AV form or deployment model into “good” or “bad,” though does occasionally caution us on second order effects of dynamics already at play. Instead, he focuses on three core themes that will drive the industry’s evolution over the coming decades:
Specialization.
Materialization.
Financialization.
As he clearly states when introducing them, these themes are sequential and therefore involve decreasing confidence and greater potential for variability as he forecasts their impacts. His first theme, specialization, addresses the already-flourishing diversity of AV platforms. It’s not just about the car or the class 8 truck, but the sidewalk delivery robot, automated micromobility devices, and transit vehicles. He also introduces “civic caravans” and “urban ushers,” the former of which amount to mobile buildings providing government services (i.e. kiosks and clinics of sorts), and the latter which perform public safety and traffic management functions (i.e. neighborhood watch or crossing guards). His broader point is that we should expect the underlying components of autonomy to diffuse into myriad platforms serving diverse functions, even if we can’t accurately pinpoint the “what” and “why” at this current juncture.
His second chapter under the specialization theme (borrowing his “Reprogramming Mobility” title) surveys more conventional mobility topics, including transportation network companies (TNC), mobility-as-a-service (MaaS), and microtransit, with a balanced look at each. While these topics are, frankly, beaten to death by academics at this point, he manages to keep the narrative fresh and offers compelling history.
The next two chapters address his second theme, materialization, which postulates we’re going to see far more attention directed to moving things and not people. This picks up the story of Amazon’s Prime delivery service and the staggering growth of e-commerce altogether. Townsend’s focus is not Interstate freight networks, but rather the last mile(s) from fulfillment centers to your door. He argues the three key enabling factors for this automated goods movement are 1) rightsizing vehicles that don’t need to account for passenger comfort, 2) repurposing existing capacity (i.e. renting out your vehicle when not in use), and 3) all-hours movement.
The first dynamic draws on what we’re already seeing with purpose-built good-delivery vehicles like those built by Nuro and Starship, and it’s a good bet that we’ll continue to see those proliferate. The second is often (annoyingly) promoted by mobility tech utopians enamored with the prospect of perfect asset utilization. I don’t discount its viability, but remain skeptical. We certainly have seen this succeed in freight transportation, and the idea of using the national fleet of 500,000 school buses to deliver goods in their downtime seems less far-fetched after I watched Atlanta Public Schools repurpose its 400+ buses to deliver tens of thousands of meals daily during the early COVID-19 lockdown. Finally, provided a secure depository for goods, there’s no reason we shouldn’t see more deliveries shift around the clock. If high asset utilization is the game, running devices around the clock will grow consumer demand and ensure an effective return on investment.
Townsend does dive into the social impacts of AV-driven growth in material demand, concerning himself with increasingly wasteful consumption habits and their environmental impacts, and labor displacement. But as with the rest of his narrative, this isn’t a one-sided affair. If anything, he appears more optimistic AVs will enable more efficient resource allocation and recycling, creating circular economies. And I find more agreement with him on labor concerns, where automation is more likely to augment drivers on existing vehicle platforms than replace them. Industrial roles (e.g. drayage, forklift) are the most vulnerable to job losses, but long-haul and delivery drivers look to be safe for the foreseeable future.
His final chapters veer into much more uncertain territory, but this endgame is where Townsend shines. He pulls together pieces he’s introduced over the book’s first two parts to speculate how financialization of AVs might reshape urban fabrics and society at-large. After tracing congestion pricing’s history to mid-20th Century America, he offers a skeptical view (one that I share) of the power it grants governments and domineering private sector players to exert over urban mobility. He describes SoftBank, Uber, and their global subsidiaries as this generation’s would-be “traction monopolies,” dominating critical infrastructure markets to the extent they exert influence over municipal governments, and are difficult, if not impossible, to dislodge locally. After asking some finance/economics friends who would know such things (they didn’t), I found the obscure term is used almost exclusively in academic literature about late 19th Century electric streetcar systems (the example Townsend uses to introduce the topic).
Basically, traction monopolies are what we know today as regulated utilities, but without the the regulation. Or before the regulation. Instead, these firms use deep reservoirs of capital and political connections to outmaneuver and starve their competition. And once firmly entrenched, these firms can bully cities into submission by choking off service at inconvenient times (e.g. what Uber and Lyft did to Austin in 2016) or use political influence to stack the deck against would-be competitors. Townsend shares my concern that congestion pricing in a place like New York City ultimately favors Uber because it has deeper pockets than competitors and can more effectively underprice its service to crush them. But he goes further and raises the prospect of a city dependent on revenues from its congestion pricing scheme, with Uber able to starve (or gorge) that city’s finances depending on its treatment by public officials. Governments may stumble into conflicts of interest with pension systems invested in potentially leveraged financial products built off projected revenues from these schemes. Automation may exacerbate this dynamic, in terms of cost and operational efficiencies enabling greater scale and winner-take-all dynamics.
We then get to Townsend’s last big idea, around the way AVs will reshape land use and urban form. In some respects, this is the most familiar territory in his entire book, as countless other urban planners/designers and futurists have been giving TED Talk-style presentations on the topic for years. I do find Townsend’s imagining of the new urban landscape to be better-conceived than most others I’ve seen, hence my fondness for his original “Reprogamming Mobility” scenarios paper. He organizes the city into four segments. The urban core, the fulfillment zone, microsprawl, and desakota (an amalgamation of the Indonesian words for “village” and city”). It’s all fairly self-explanatory with the exception of the desakota, which is effectively what planners refer to as suburban edgeless cities or exurbia.
Urban cores remain the most attractive real estate, with materialization enabling much cheaper and instant access to [stuff]. Progressive transportation policies facilitating active and more satisfying lifestyles also play a role. Fulfillment zones sandwiched between cores and suburbia (microsprawl) are not a new idea, but are the most transformative segment in Townsend’s hierarchy. He aptly describes them as “beachheads” for same-day delivery into the urban core. Today most fulfillment centers are sited in suburbia or even exurbia, where land is cheap and massive facilities can easily exceed a million square feet. With right-sized AVs and generally more efficient transportation systems, you could conceivably process enough volume to justify facilities closer to the urban core, capable of instant on-demand delivery. Microsprawl is less a negative than most planners fear, as Townsend envisions ways in which AVs can enable more effective transit systems and efficient lifestyles. Finally, desakotas are exactly what you’d expect them to be, resembling the same bizarre mix of exurban residential, industrial, and agricultural uses you see today.
The book closes with two chapters that are Townsend’s most prescriptive. He charts an evolutionary path for transit operators, regulatory responses to curb Amazon’s power, and new measures to monetize the curb in lieu of declining parking and traffic fine revenues. Though on the latter, he raises the same concerns over traction monopolies he did previously with ridehailing. He also emphasizes the extent to which our AV future will be highly iterative, with setbacks, redirections, and policy shifts shaping commercialization and scaling of these technologies. I’ve recently been thinking about one of the key differences between this transportation revolution and those before it, which is our mature policy and legal environment. Even as we’ve allowed Uber, Tesla, Bird and others to grossly abuse public trust with occasionally disastrous consequences over the past decade, this is not early automobility at the turn of the 20th Century, or the Wild West in which railroads emerged. Just as tech giants have twisted legislators to their advantage over the past 20 years, we can’t rule out a paradigm shift against them (though a cynic might). Regulatory and legislative battles in Washington, D.C. will shape AV deployments in unpredictable ways, as action to address climate change, inequality, or even foreign adversaries grows.
Ghost Road is an imaginative guide for anyone trying to think through the next decade, and beyond. It truly one of the better books I’ve read of any genre over the past several years, and certainly the best on vehicle automation and/or urban planning. Highly recommended.
A couple weeks ago, and prior to picking up Ghost Road, I had put together some thoughts on Twitter about the shifting focus from robotaxis to more narrowly defined AV use cases. Since I think much of it dovetails nicely with Ghost Road, I’ll elaborate here.
In retrospect, automating on-demand, point-to-point ridehailing was the hardest market to address for three reasons:
First, even with some degree of geofencing, it’s just incredibly hard to develop automated vehicles that can operate in all environments at all times of day as reliably as a human driver. In theory, AV ridehailing development is currently focused on Phoenix and its suburbs because of its relative ease. Dry, hot environment, flat terrain, low density development with wide, gridded streets and relatively few pedestrians or bicyclists to navigate. Some AV firms have started with Pittsburgh or San Francisco under the logic that if they can tackle the hardest environments first, they’ll be able to scale across the country much faster. I see the “low-hanging fruit approach” as more viable, but neither has strong commercial prospects in the short- to medium-term because, even within a metro like Phoenix, scenario variability will likely keep ubiquity at arm’s-length.
Second, moving people is harder than moving things. This is where Townsend’s materialization theme nails the industry’s likely trajectory. Most deliveries aren’t time-sensitive, and those that are can suffer delays on the order of hours without prompting customers to seek more timely alternatives. A ten minute delay on your burrito is less aggravating than a ten minute delay en route to meet a client, catch a flight, or rendezvous with a date. Although a delayed burrito might be soggy, and therefore hot meals are more challenging than a grocery or clothing shipment. Passengers will expect a smoother ride, and simultaneously less deference to other vehicles and greater safety than goods. The cost of a ride may need to be significantly lower than a human-driven alternative to coax passengers into accepting a slower autonomous alternative. (I remain skeptical of AV ridehailing economics.)
Finally, returns on partial automation are lowest — and possibly negligible — with ridehailing. Uber will not both invest in SAE level 2 or level 3 systems for fleets it owns and continue to pay drivers. Those features alone don’t make Uber’s service more competitive, and the insurance savings from fewer theoretical collisions just won’t be that significant to warrant the added hardware and software costs. It’s basically all or nothing. However, lower levels of automation or more restricted level 4 operational design domains offer significant potential cost savings for fleets in freight, urban delivery, transit, warehouse, or other off-road use cases. Maybe in the short-term you can’t pull the driver out of a semi, but you can automate certain yard functions and easier long-haul segments, or add assistive features that offer fuel savings, reduce liability, and enable policy shifts on regulations (e.g. drivers’ daily/weekly hours). Townsend addresses some benefits for fixed-route transit bus partial automation that are much more viable in the short-term. And then much of his book examines purpose-built AVs or modifications of standard platforms for off-road scenarios (e.g. warehouse, terminal, etc.).
Perhaps it was necessary to ignite the hype machine with dreams of ubiquitous autonomous Ubers to catalyze the wider industry. The US military has been working on vehicle automation for more than two decades, and warehouse robots (e.g. Amazon’s Kiva) were a thing before AV-mania struck in 2015-2016. But the rapid advances we’ve seen in on-board computing, sensing hardware (particularly lidar), and other complementary automotive software and hardware components were triggered by the mad rush to develop fleets of test vehicles, and “win” a perceived race to deploy commercial robotaxi service. However, after billions of capital flooded into these ventures with little immediate prospect for financial returns, investors have clearly turned their attention to commercial use cases.
That’s not to say light passenger vehicle automation has been put on hold, after all, in just the past few months Amazon just acquired Zoox for north of $1 billion, Volkswagen closed a massive, multibillion dollar investment into Argo.ai, and other traditional automakers have continued sinking millions into their existing AV plays. But momentum in the industry has clearly shifted toward moving things, and not people. Even industry-leader Waymo has started soliciting outside investment to staunch its autonomous ridehailing cash-burn, as it grows its automated trucking operations. When you contemplate the hundreds of millions of dollars it will cost to deploy an autonomous ridehailing network in the US, with no guarantee of success against human-driven alternatives, you start to wonder when investors will have the stomach for another protracted speculative battle. It may not be until China successfully demonstrates a viable commercial robotaxi network (which can more easily control road environments) that we see another mad rush from investors in the US market.
Since this might be my last AV post here for a few months (I have a lot to write about Atlanta transit issues!), I wanted to share a couple important professional updates here as well that tie into the Ghost Road discussion.
- Back in July of last year, CTE was pulled into a proposal opportunity with the Connecticut Department of Transportation, New Flyer, and Robotic Research for the Federal Transit Administration’s (FTA) Integrated Mobility Innovation (IMI) grant. I flew to Hartford on a moment’s notice, rode the CTfastrak bus rapid transit line, my first time ever on any BRT (yes, it was impressive), and ultimately co-wrote a proposal to automate multiple 40-foot New Flyer battery electric buses on it. Well flash-forward to March…we won the grant award, and assuming all goes according to plan, the first automated transit buses in North America should enter revenue service on the CTfastrak in 2022. I’m thrilled to be supporting the project, in what promises to be a milestone for the US transit industry. We’re limiting development to SAE level 3 automation, where a trained operator will be remain behind the wheel at all times. Though we expect the automated driving systems should be able to drive the buses without any human intervention on the CTfastrak, which is a completely separated guideway, the route does include a mixed-traffic loop in Downtown Hartford where operators will need to resume full control of their vehicles. We will be focusing on demonstrating automated precision docking at station platforms and platooning up to three buses together, in addition to general automated driving. We will also be looking at driver and passenger acceptance of the technology, as well as net energy impacts from the automated driving system. We’re stressing this remains a level 3 system requiring a fully-trained driver, as this is the most viable long-term path for the technology’s adoption both technically and politically. Which leads to my next point…
- Since fall of last year, I’ve been working with Congressman Dan Lipinski’s office (supported by my leadership and our government affairs representation) to come up with a transit bus automation program for inclusion in the next surface transportation authorization legislation. We settled on a structure akin to the National Fuel Cell Bus Program passed under the 2005 SAFETEA-LU legislation because that model proved successful in catalyzing the entire zero-emission bus industry. It launched a new domestic manufacturer (Proterra), accelerated electric drivetrain development on buses (which also stimulated investment into battery electric alternatives), and supported growth of the wider US hydrogen transportation industry. Lo and behold, “SEC. 2805. NATIONAL ADVANCED TECHNOLOGY TRANSIT BUS DEVELOPMENT PROGRAM.” is now part of the INVEST in America Act. This new program would provide a dedicated federal funding stream to develop automated bus and supporting technologies up to SAE level 3 automation, with a focus on safety and driver quality of life improvements. While those parameters may seem limited, that’s deliberate. Simply put, there is no future for transit bus automation without buy-in from organized labor (which did endorse the legislative language). The industry must be proactive about working with labor to find a development path that simultaneously ensures transit buses evolve to remain competitive with competing modes, and provides significant operating benefits without cutting pay or jobs. The potential gains are there, but federal funding is necessary to develop these capabilities. Development is just too expensive for individual bus manufacturers or sponsoring transit agencies to bear without federal support.
- I’ve been writing (for work) and speaking a lot over the past year on the challenges and benefits of transit bus automation, but have yet to publish on it. I probably need to just write something for this blog or another publication. But if you want to learn more before then, my appearance on the Mobility Podcast earlier this year involved a good bit of discussion on the topic.
Nathaniel Horadam 