October 5, 2017

Limiting freeways to autonomous vehicles is a bad idea (today)

Can a good idea implemented too soon actually be a bad idea? I would say so, if the early implementation imposes costs in return for little benefit.

About a year ago I posted about an under-developed proposal from Madrona Venture Group to restrict I-5 between Seattle and Vancouver to autonomous cars only. The Madrona team is back with an update to their proposal, which addresses some of its flaws but still comes up short.

The first issue is timing. The revised proposal recommends making I-5 autonomous-only “by 2040.” The new suggestion is a lot more reasonable than their previous recommendation of “ten to fifteen years,” but still ignores decades of evidence about the speed of technology adoption in the transportation sector. As I showed in my post last year, 20 years from initial market introduction is a fairly aggressive schedule for all new cars to become autonomous. But the average car in the U.S. is 11 years old, and 60% of cars in the US survive 15 years or longer. This means that even when the autonomous transition is complete among new cars, manually driven cars will be in circulation for decades more. To their credit, the Madrona team recommends keeping I-5 open to manually driven cars at off-peak hours, which is a sensible and nearly costless measure. Still, closing the freeway to drivers in these vehicles will disproportionately impact lower-income travelers who cannot afford a new car.

Talking heads love to roll their eyes at those dinosaurs in the auto industry, and hold up smartphones as evidence that adoption rates are so much faster for the technologies sent down to us from the Gods of Olympus Silicon Valley. But the price of a smartphone is 1-2 orders of magnitude less than that of a car, and the service life of a phone is also an order of magnitude shorter than that of a car. Forget apples and oranges; this is apples and lawnmowers. To understand the pace of transitions in the transportation sector, consider tech darlings Uber and Lyft. By all accounts resounding successes, five years after Uber’s debut they nevertheless accounted for just 2% of all trips. In San Francisco.

Now, the above discussions of transition time all measure from the time of market introduction, but the Madrona team is also overly optimistic about dates of introduction. They write,

What does it take to get to this future? On the technology side, not much. We will likely have the technology to operate fully autonomous cars (Level 5) in the next 3 years, especially for well-mapped, frequently traveled roads, with known weather and surface conditions (such as I-5). Tesla claims its most advanced system will be capable of Level 5 later this year.

First of all, a Level 5 car that is limited to “well-mapped, frequently traveled roads, with known weather and surface conditions” is more accurately known as a Level 4 car. That is literally the definitional difference between Level 4 and Level 5.

But more importantly, Tesla has said no such thing. Tesla is projecting two years for full automation. And without diminishing the company’s many great achievements, it is fair to say that hitting deadlines is not exactly their forte.

The authors also cite the autonomous Audi A8 as nearer-term evidence of market readiness, but the Audi is most notable for its human-machine interface, not because it is driverless. This becomes problematic when viewed alongside Madrona’s evidence-free assertion that “These vehicles will likely to be part of ride-sharing fleets and other travel services that will benefit the entire population and lower the cost of transportation for everyone.” Shared mobility fleets need Level 4-5 automation, since driverless operation is a necessity if a vehicle is to deadhead from one dropoff to its next pickup.

Finally, the revised report still doesn’t address the basic question of Why? Why is this policy needed today, and what will it accomplish? If AVs are providing shared rides, then by all means open up HOV lanes to them. But what is the justification for turning over scarce right-of-way to these vehicles when carrying a single occupant? If AV technology is half as good as we all expect, adoption will happen without this additional government incentive. If the goal is to increase capacity because AVs work best with other AVs, then why should this policy precede the adoption of a critical mass of AVs?

None of this would matter if the policy didn’t carry significant downsides. But automation can increase demand for travel, leading to more congestion, more energy consumption, and more pollution. Moreover, an influx of AVs could degrade the performance of HOV lanes, further eroding incentives for commuters to share rides.

I stand by my judgment that there will come a time when it makes sense to limit some or all public roads to AVs, but trying to prescribe that timeline today is a fool’s errand. When that time arrives, it will be fairly obvious to most of us: most people will be using AVs for most of their travel, and we will have data establishing that AVs can more effectively improve safety, capacity, and efficiency when operating an an AV-only environment. In the meantime, a less charitable wag might conclude: “VCs support policy to let them drive their Teslas in the HOV lane.”