8 Comments

Arguably, in the long run, services like Waymo will not be competing with Uber and Lyft, they will be competing with private vehicle ownership. I was curious, so I asked OpenAI Deep Research for a report on usage levels of passenger vehicles overall (https://chatgpt.com/share/67eea043-9be4-800b-9a9a-2ef70ce9ef7e). The results indicate that passenger vehicles in the US are overprovisioned by roughly 10x vs. peak usage. Thus, if Waymo were provisioned for 1x peak usage, it would get roughly 10x better utilization than today's overall vehicle fleet.

I think I may be saying more or less the same thing as Ethan's sibling comment, just coming at it from a different direction – the fact that the status quo of private vehicle ownership is economically viable implies that the need to serve peak usage periods should not push costs out of bounds.

Of course, I'm glossing over all sorts of details here, but in the big picture it seems like it should be economically feasible to provision a Waymo-like service for peak usage, once the cost of the AV hardware comes down a bit? Though I could imagine that in the current early-adoption period the peak-to-mean ratio for Waymo is higher than for passenger vehicles overall (e.g. relatively high usage during events such as a big concert, relatively low usage during routine commute hours).

(Caveat, I made zero attempt to investigate the figures Deep Research produced, so they could be way off.)

Expand full comment

As I replied to Ethan elsewhere, scaling to peak demand is something that, absent other constraint, robotaxi companies would do. The financial upside is high and the downside is low.

But they ARE otherwise constrained… politically.

What do you do with redundant vehicles during the off-peak? Why, you have them cruising the streets, so that customers who would otherwise have to wait 8 minutes for a pickup instead wait 4. Or you have ‘happy hour’ pricing during the off-peak to turn latent demand into actual demand.

But the side effect of THIS is greater congestion during the off-peak, which annoys human drivers, delivery companies, and so forth, and to whom elected officials respond. The result is vehicle caps, akin to the fixed number of taxi medallions in big cities last century.

That to me is the likeliest outcome. A BETTER outcome would be congestion charges in all big metros… but I’m not just a policy technocrat, I’m also a realist.

Expand full comment

Well, in any world where robotaxis are scaling to any significant fraction of overall passenger transportation, I presume there'd be a significant decrease in private vehicle ownership: the robotaxis will be displacing other forms of transportation, including private vehicles, yes? This will free up parking spots, which the taxis can then use when not in demand. Some of those displaced private vehicles will have been living in garages, which won't be available to the robotaxis (barring some unlikely-seeming rent-your-garage scheme), but some will have been parked on the street or in other accessible locations. Given that one robotaxi should be able to displace many private vehicles, I would guess that the overall balance should be to free up street parking?

Or, taking a step back: robotaxis seem to be fundamentally much more efficient than private vehicles in terms of total fleet size required. Also they should be better drivers, and more flexible, conducive to interoperating with public transit, ride sharing, etc. A world with more robotaxis and fewer private vehicles ought to allow our streets and parking to be less congested, not more. Achieving this positive outcome will require appropriate incentives and city planning, but the fundamental efficiencies seem so strong that I'd think the design challenges won't be extreme – a "merely decent" set of policies ought to yield a positive outcome, we'll be sailing with a strong tailwind.

Expand full comment

Steve, if you're thinking along these lines, and want to read a rigorous deep-dive into the relative merits and barriers of different scenarios in which automated driving diffuses, have I got a book for you...

Expand full comment

>>What do you do with redundant vehicles during the off-peak? Why, you have them cruising the streets, so that customers who would otherwise have to wait 8 minutes for a pickup instead wait 4. <<

I see no reason they need to be movibg around when unoccupied. They just need to be spread around in a smart way. Yes they need a parking place, but parking unoccupied driverless vehicles must be a much smaller problem than parking all the vehicles for the commuters into the city.

Let's imagine that the goal is five minute response time to a call. As a potential rider, 5 minutes seems quite reasonable to me. Now Waymo (or whoever) needs to have enough unoccupied vehicles that even at peak demand there is still at least one waiting vehicle within 5 minutes of every point in the city.

I can imagine Waymo buying a city lot that is vacant or covered with less valuable real estate. They excavate the lot down about 12 ft and build a basement with a ramp to the surface. Then pour a concrete roof over the basement and build a commercial establishment on top.

>>The number of cars that can park in a city lot depends on the lot's size, layout, and whether it's designed for efficient parking or includes landscaping and walkways. A standard acre can hold roughly 144 traditional parking spaces, but practical layouts typically accommodate 100-115 spaces. <<

Rounding down, lets say 100 spaces for driverless cars. Since the vehicles are interchangeable, they can be parked bumper to bumper. Everytime leave vehicle is activated, the other vehicles in the queue move up a space. So 100 vehicles would actually require much less than an acre.

Now Waymo needs one of these basements within a 5 minute drive of every point of the city.

>>During rush hour, city traffic speeds typically drop significantly, with average speeds ranging from around 10 to 20 miles per hour, depending on the city and time of day.<<

We definitely need to be conservative in this case, so 12 miles per hour. One mile every 5 minutes.

Now we need a grid of basement parking lots every mile in two dimensions. I think that works out to one parking lot per square mile. Imagine a city of 100 square miles. That's 100 parking lots and 10,000 vehicles to provide 5-minute response time.

Quite a bit of slack there, you can go from 100 parked vehicles to 1 parked vehicle and still provide 5-minute response time. The underground parking means zero usurpation of commercial/living/recreation space.

Sounds good to me.

Expand full comment

I agree that we'll continue to see human rideshare drivers serving times of peak demand (and even off-peak due to personal preferences) for a long time.

But I do think robotaxi companies are likely to scale their fleets towards serving peak demand. Fully utilized vehicles would end up driving enough to get past their useful lifetime of 300k miles within a couple of years. Scaling for peak demand would just spread that usage out over a longer period of time, with fairly minimal additional costs (e.g. upfront investment is not recovered as quickly + lot space).

I feel like this point is under-discussed, but I allude to it in this footnote: https://www.2120insights.com/p/how-autonomous-vehicles-will-change#footnote-9-152869747

Expand full comment
Comment deleted
1d
Comment deleted
Expand full comment

The problem is political consequences. Scaling to peak demand is something that, absent other constraint, robotaxi companies would do. The financial upside is high and the downside is low.

But they ARE otherwise constrained… politically.

What do you do with redundant vehicles during the off-peak? Why, you have them cruising the streets, so that customers who would otherwise have to wait 8 minutes for a pickup instead wait 4. Or you have ‘happy hour’ pricing during the off-peak to turn latent demand into actual demand.

But the side effect of THIS is greater congestion during the off-peak, which annoys human drivers, delivery companies, and so forth, and to whom elected officials respond. The result is vehicle caps, akin to the fixed number of taxi medallions in big cities last century.

That to me is the likeliest outcome. A BETTER outcome would be congestion charges in all big metros… but I’m not just a policy technocrat, I’m also a realist.

Expand full comment

>>What do you do with redundant vehicles during the off-peak? Why, you have them cruising the streets, so that customers who would otherwise have to wait 8 minutes for a pickup instead wait 4. <<

I see no reason why they need to be moving around when unoccupied. They just need to be spread around the city in a smart way. Yes they need a parking place, but parking unoccupied driverless vehicles must be a much smaller problem than parking all the vehicles for the commuters into the city.

Let's imagine that the goal is five-minute response time to a call. As a potential rider, 5 minutes seems quite reasonable to me. Now Waymo (or whoever) needs to have enough unoccupied vehicles that even at peak demand there is still at least one waiting vehicle within 5 minutes of every point in the city.

I can imagine Waymo buying a city lot that is vacant or covered with less valuable real estate. They excavate the lot down about 12 ft and build a basement with a ramp to the surface. Then pour a concrete roof over the basement and build a commercial establishment on top (or a park etc.).

>>The number of cars that can park in a city lot depends on the lot's size, layout, and whether it's designed for efficient parking or includes landscaping and walkways. A standard acre can hold roughly 144 traditional parking spaces, but practical layouts typically accommodate 100-115 spaces. <<

Rounding down, lets say 100 spaces for driverless cars (DVs). Since the vehicles are interchangeable, they can be parked bumper to bumper. Every time the lead vehicle is activated, the other vehicles in the queue move up a space. So 100 vehicles would actually require much less than an acre.

Now Waymo needs one of these basements within a 5 minute drive of every point of the city.

>>During rush hour, city traffic speeds typically drop significantly, with average speeds ranging from around 10 to 20 miles per hour, depending on the city and time of day.<<

We definitely need to be conservative in this case, so 12 miles per hour. One mile every 5 minutes.

Now we need a grid of basement parking lots every two miles in two dimensions. I think that works out to one parking lot per 4 square miles. The underground parking means zero usurpation of commercial/living/recreation space.

Imagine a city of 100 square miles. That's 25 parking lots (@ $10 million per) and 2500 DVs (@ $100,000 per) to provide 5-minute response time. ($500 million total -- cheap!)

Quite a bit of slack there: you can go from 100 parked vehicles to 1 parked vehicle and still provide 5-minute response time. Of course at rush hour, riders are constantly exiting DVs and those vehicles can also respond to calls. So the average time would be less than 5 minutes. Also note that when a DV is not called for, it goes to the nearest parking basement. No need to return to where it began.

Sounds good to me.

Expand full comment