Assessing Cavnue’s big bet on AV infrastructure

With tens billions of dollars pouring into the space over the past decade, we’ve seen dozens of connected and automated vehicle (CAV) testbeds and regional initiatives emerge across the country. Some of these are purely economic development plays, without a research institution or major technology developer anchoring them; others are university-funded (often grant-aided) testbeds leveraging some core competency of that institution. Then there are the active deployment/test sites for major developers. Think Waymo, Ford/Argo AI, GM Cruise, the other automotive OEMs, Zoox, Nuro, and the whole crew of low-speed automated shuttle developers.

Enter Cavnue, a program of venture-backed Sidewalk Infrastructure Partners (Google subsidiary) that aims to build dedicated CAV infrastructure across the country via private-public partnerships, starting with a pilot corridor between Detroit and Ann Arbor, MI. The corporate structure and nexus of private and public entities involved is complicated, but that pilot project rolled out in mid-2020 with backing from the State of Michigan and automotive industry heavy hitters. Cavnue is interesting because it represents both a clear statewide economic development initiative for Michigan and a theoretically replicable business model for dedicated CAV infrastructure.

Rendering of Cavnue’s dedicated AV lanes traveling between Detroit and Ann Arbor, MI.
(Credit: Cavnue)

I want to talk here primarily about the model of dedicated CAV lanes, and not broader economic development objectives and outcomes. I’ve grown increasingly disillusioned with AV-as-economic-development projects, as most have failed to produce meaningful successes (even if government sponsors rarely admit it), but it’s a more complicated question and one I’m not equipped to tackle. Michigan’s natural advantage as the US automotive capital and its ability to assemble major industry players raises the likelihood of ecosystem benefits from this pilot, though the American Center of Mobility’s saga remains a cautionary tale.

I’m also going to avoid opening the connected vehicle (CV) can of worms, since that topic is already the subject of a worn-out debate and multiyear national policy fight. So for the purposes of this piece, I’m sticking strictly to automated vehicle (AV) technologies, though CV infrastructure may ultimately add to costs and dependencies.

Why build AV-only lanes?

For the most part, major technology developments have focused their testing activities in the easiest environments, namely low-density Sun Belt metros and Interstate corridors. While lax (and clear) regulations influence these decisions, the combination of warmer weather (i.e. no snow) and fewer other road users suggests these states/regions will facilitate earlier commercial deployments. Some developers have chosen more challenging environments like the urban cores of San Francisco and Pittsburgh, with the belief that tackling more complex scenarios will accelerate technology development. I lean toward the former approach, but there’s at least some merit in the latter.

The basic idea around building dedicated infrastructure for AVs involves limiting interactions with human-driven vehicles and other road users to reduce operating complexity. In technical terms, this means a highly-constrained operational design domain (ODD), or the conditions in which an AV needs to operate. And an easier operating environment likely means an earlier commercial deployment, provided a viable business case. So there’s a “build it and they will come” mindset, where these facilities make specific corridors more attractive for technology developers. While program leadership uses some language around testing and moving AV technology forward, Cavnue’s Detroit-to-Ann Arbor project clearly aims to set that corridor in the vanguard of commercial deployments.

Is there value in AV-only lanes?

Aside from those focusing on off-road use cases (e.g. ports, warehouses, and other terminals), most AV technology developers are aiming to deploy their vehicles on roads without the need for any specific supporting infrastructure. New infrastructure is expensive, requires some government to plan and pay for it, and, if prior “transportation revolutions” are any indication, may not see standardization for decades. Aside from those that get government contracts to plan, design, and build this infrastructure (e.g., engineering consultants), private firms do not have the time or investor patience to wait for it.

And here’s the thing about dedicating lanes to automated vehicles: the benefits of doing so vary greatly by mode and purpose. Think of them like a conveyor belt in a factory, automatically delivering goods from one station to another. If you have a high volume of goods to transport from one end to points along it, and the time or labor costs of getting to/from the conveyor belt are low, then that investment is worthwhile. You can further automate that movement if the additional cost is worthwhile. Similarly, the distance and complexity between any dedicated AV lane and an origin or destination will dictate its value.

The value of a dedicated lane for automated trucks differs from that of one for automated buses, robotaxis (i.e., autonomous Ubers), or delivery vans because that last-mile experience to their passengers’ or goods’ destinations differ greatly.

I’ll address them one-by-one:

Would Cavnue accelerate commercial robotaxi deployments? No. Beyond my general skepticism of commercial viability for robotaxi service, this is the least valuable use case for a dedicated AV lane. What good is a dedicated lane between two cities if the automated vehicle ODD doesn’t cover a sufficiently large swath of both cities to make point-to-point automated travel viable? Even from a demonstration standpoint, there isn’t much value in proving a fleet of light vehicles can safely travel in a straight line for 40 miles, completely separated from human drivers. That was a decades-long dream partially-realized in 1997 with the Automated Highway System program’s San Diego demonstration, but we are well past that as an industry.

Would Cavnue accelerate commercial AV truck deployments? No. Let’s go back to the conveyor belt framework. Building a dedicated lane for automated trucks to travel between Detroit and Ann Arbor only facilitates automated driving while in that lane. Unless the lane starts and ends near freight terminals, or has them dotted along the route with dedicated facilities to access/exit, these trucks will need to spend considerable amounts of time navigating that remaining distance in challenging road scenarios. Similarly, freight trains require spurs to navigate industrial zones for “doorstep” delivery, and high-volume intermodal terminals require buildout of a sufficient last-mile distribution network. A 40-mile corridor that still requires commercially-licensed drivers to navigate several miles to and from the dedicated facility is not a great value proposition for an early deployment. Especially because Detroit to Ann Arbor is not a high-volume truck freight corridor, as most of the flows to and from Detroit travel via Toledo to the south.

An Embark automated truck operating in Nevada. Embark is among the leading automated trucking startups, and recently went public via $5.2 billion SPAC deal. (Credit: Embark)

We’re already seeing the future of automated trucking cluster around Southwest metros such as Dallas, Phoenix, and Los Angeles, where long stretches of dry Interstate highways raise the value of a hypothetical conveyor belt. And these corridors are not developing dedicated infrastructure, meaning AV technology for trucking is likely to evolve in a manner that doesn’t depend on it. From there, on-road trucking automation will focus on other high-volume corridors or those where labor availability is strained by long, desolate distances. For instance, paying a short-haul driver to cover the final few miles to a distribution center after a truck’s automated 1000-mile journey between Dallas and Phoenix is a strong early business case. Commercial entities will choose these sorts of longer-haul routes for automated trucking operations before opting for short intercity routes in challenging urbanized environments. That presents a high likelihood Cavnue’s infrastructure is outmoded before it’s useful.

Would Cavnue accelerate commercial AV delivery van deployments? No. And for similar reasons as those for robotaxis. If you need the ability to directly reach your customers, you will have to master more complicated urban environment before less challenging intercity corridors.

Would Cavnue accelerate commercial automated transit bus deployments? Yes! And to be fair to Cavnue, it led with the transit automation use case in its launch announcement, and has promoted that dimension in subsequent industry forums. The reason dedicated lanes for automated buses makes sense is because dedicated lanes for human-driven buses makes sense. We see this in numerous bus rapid transit (BRT) projects across the United States, and it is standard for bus routes elsewhere in the developed world. These projects string together dense commercial and residential nodes with high-capacity vehicles (40+ passengers), which passengers typically access by walking, biking, or connecting via another transit line. Aside from trips from and to the bus depot, vehicles never have to leave the dedicated lane (or terminal station with controlled access).

One could question whether there’s value in messing with automation at all, rather than just building standard bus-only lanes. I maintain transit automation is worthwhile even if a driver needs to remain on board to drive from/to the depot or in the event of an emergency, and for passenger safety and assistance. While I’m long overdue for a longer writeup, I did explain the value proposition on my recent Autonocast appearance and in an earlier Twitter thread. (I am currently supporting the first US automated bus demonstration for the Connecticut Department of Transportation’s CTfastrak deployment in Hartford, alongside New Flyer and Robotic Research.)

Connecticut Department of Transportation’s CTfastrak dedicated busway. (Credit: CTDOT)

Though we can always improve last-mile transportation, shortcomings in that network rarely doom transit projects due to lower customer expectations (sadly) and different operating economics. While total return on investment does matter, transit projects aren’t evaluated on an individual project’s financials once delivered. The system as a whole just needs to stay “in the black” with a combination of fare revenues and local, state, and federal funding support.

The challenge of Cavnue as an automated bus project is that questionable project profitability, hence why its leadership is promoting a shared facility for AVs of all transportation modes. Unless the objective is to sell turnkey fully integrated CAV lanes for transit use, which may have quite a few interested customers, these facilities will not be profitable without opening them to other vehicles. And I just don’t see the value proposition for trucks, delivery vans, and robotaxis.


I’m watching Cavnue closely because it has assembled serious industry support alongside key government sponsors, and frankly any Google-backed initiative in the physical world is worth watching (mixed track record notwithstanding). That said, Cavnue’s ambitious framework is built on an outdated premise, albeit one still promoted by civil engineers with a vested interest in infrastructure dependence for the AV technology.

That doesn’t mean the Michigan project will be failure. As I’ve explained, there is real value in building out dedicated transit lanes, and automated buses will certainly launch in these types of settings first. I just don’t see the value in multi-purpose AV lanes, even if the service impact of allowing access to private vehicles is negligible. Sidewalk Infrastructure Partners may have other ideas about how Cavnue can generate enough revenue to meet investor expectations, but Michigan’s project sponsors need to be clear-eyed about what the project can actually accomplish. The same goes for governments in other geographies targeted for Cavnue deployments.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s