All else being equal, I think it is almost always preferable (from a societal perspective) for an empty AV to be parked rather than cruising. In both cases, the vehicle is occupying valuable public space. However, the cruising vehicle consumes more energy, emits more pollutants, occupies more space, and increases the risk of collisions for other road users.
To your point about curb space being scarce in business districts, when (and where) curb space is scarce, road space is also scarce, except for around 10-11:15am when demand for AV ridehailing in business districts is also minimal.
Looking at it from another perspective: AV fleet operators are definitely repositioning their vehicles to meet anticipated demand. Because they don't internalize the costs of emitting pollutants or occupying public space (but do internalize energy costs and the risk of collisions), and they don't have to pay for road use, I would expect them to spend more of their P1 time cruising/repositioning than the social optimum.
We recently did a related study, funded in part by Waymo. We analyzed how curb parking restrictions for SAVs in SF would affect VMT, curb usage, customer wait time, and request matching rate. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5526419
Absolutely agree with you about the need to do this same kind of analysis for TNCs with an expanded lens to account for driver wellbeing. The issue is that the CPUC has still not released comparable data. It had been five years now of no public reporting… Perhaps another state makes sense data available?
“The issue is that the CPUC has still not released comparable data. It had been five years now of no public reporting… Perhaps another state makes sense data available?”
This was a great read, and thanks for sharing the Google Sheet with the CPUC data you processed for this analysis. The footnotes were really thoughtful too.
I've been playing with your dataset looking at the distance in P1, P2, and P3 per trip. The average trip distance (P3) has increased consistently, but the average P1 and P2 per trip hasn't been so consistently directional. That makes me think the improvement in P3's share of total VMT is driven more by trips getting longer than by Waymo doing less repositioning per trip.
The shift from parked to moving during P1 makes sense through that lens too. If they're still driving similar distances to reposition but have way less time between trips, the parked portion naturally gets squeezed out first.
Either way, the demand growth story is impressive. Going from 77k to 1M+ trips in under two years while average trip distance increases is a good problem to have. Looking forward to seeing how Zoox's and Nuro's numbers compare once they're operating commercially in California. My guess is Nuro/Uber service will be a lot higher percentage of on-trip mileage and time because of the bigger pool of requests it will serve as a non-directly-bookable part of Uber's other products initially.
"Sometimes parking may be preferable, assuming Waymos are parking in legal locations and paying for spots when at metered locations (which would require a seamless, backend curb management solution)."
Based on ISO DTS 25614 it orchestrates PUDO spots by assigning them to vehicles. The system does not care if these spots are used for passengers, goods, waiting between assignments, cleaning, maintenance, charging batteries, etc. it is also neutral to whether the spots are monetized, as that is the decision made by the manager of the spots. Hence, the spots may be on private property or at the urban curb.
Happy to read your analysis. Waymo’s progress is impressive, and their operational data is invaluable. But the numbers also highlight a deeper issue: today’s vehicle width is inherently wasteful for the way urban mobility now works. Even the most efficient AV fleet still occupies far more road and curb space than is necessary to move a single person.
The 39-inch-wide Tango EV already proves what’s possible when we rethink vehicle geometry. A tandem, one-person-wide footprint dramatically reduces the road-width each vehicle consumes—whether driving, waiting, or parking. That design advantage scales directly into lower congestion, faster trip times, and a smaller curb footprint for commuter cars, highway patrol fleets, and autonomous services.
If AV networks are going to grow from thousands to hundreds of thousands of vehicles, optimizing width isn’t optional—it’s foundational. #ThinMobility https://youtu.be/8VPz38Xkgsc?si=tE23hG1C-TecM8nM
Great work!
All else being equal, I think it is almost always preferable (from a societal perspective) for an empty AV to be parked rather than cruising. In both cases, the vehicle is occupying valuable public space. However, the cruising vehicle consumes more energy, emits more pollutants, occupies more space, and increases the risk of collisions for other road users.
To your point about curb space being scarce in business districts, when (and where) curb space is scarce, road space is also scarce, except for around 10-11:15am when demand for AV ridehailing in business districts is also minimal.
Looking at it from another perspective: AV fleet operators are definitely repositioning their vehicles to meet anticipated demand. Because they don't internalize the costs of emitting pollutants or occupying public space (but do internalize energy costs and the risk of collisions), and they don't have to pay for road use, I would expect them to spend more of their P1 time cruising/repositioning than the social optimum.
We recently did a related study, funded in part by Waymo. We analyzed how curb parking restrictions for SAVs in SF would affect VMT, curb usage, customer wait time, and request matching rate. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5526419
Great work! Now do TNCs, noting that the P1 and P2 times for every human-driven car affect the driver:
1. they are sitting
2. they are breathing air pollution
3. for P0, they are stuck not earning
4. for P1, they are risking a collision
5. their car is depreciating
Absolutely agree with you about the need to do this same kind of analysis for TNCs with an expanded lens to account for driver wellbeing. The issue is that the CPUC has still not released comparable data. It had been five years now of no public reporting… Perhaps another state makes sense data available?
“The issue is that the CPUC has still not released comparable data. It had been five years now of no public reporting… Perhaps another state makes sense data available?”
1. MN: https://www.dli.mn.gov/sites/default/files/pdf/TNC_driver_earnings_analysis_pay_standard_options_report_030824.pdf
2. Chicago, via UC Berkeley: https://laborcenter.berkeley.edu/wp-content/uploads/2025/01/Chicago-Transportation-Network-Drivers-Earnings-and-Projections.pdf
Both via Claude.ai query
NY TLC has very interesting data.
Very interesting analysis. Looking forward to a 2nd post!
This was a great read, and thanks for sharing the Google Sheet with the CPUC data you processed for this analysis. The footnotes were really thoughtful too.
I've been playing with your dataset looking at the distance in P1, P2, and P3 per trip. The average trip distance (P3) has increased consistently, but the average P1 and P2 per trip hasn't been so consistently directional. That makes me think the improvement in P3's share of total VMT is driven more by trips getting longer than by Waymo doing less repositioning per trip.
The shift from parked to moving during P1 makes sense through that lens too. If they're still driving similar distances to reposition but have way less time between trips, the parked portion naturally gets squeezed out first.
Either way, the demand growth story is impressive. Going from 77k to 1M+ trips in under two years while average trip distance increases is a good problem to have. Looking forward to seeing how Zoox's and Nuro's numbers compare once they're operating commercially in California. My guess is Nuro/Uber service will be a lot higher percentage of on-trip mileage and time because of the bigger pool of requests it will serve as a non-directly-bookable part of Uber's other products initially.
"Sometimes parking may be preferable, assuming Waymos are parking in legal locations and paying for spots when at metered locations (which would require a seamless, backend curb management solution)."
The Pudocity system addresses this: https://www.intertraffic.com/news/smart-mobility/orchestration-as-a-service
Based on ISO DTS 25614 it orchestrates PUDO spots by assigning them to vehicles. The system does not care if these spots are used for passengers, goods, waiting between assignments, cleaning, maintenance, charging batteries, etc. it is also neutral to whether the spots are monetized, as that is the decision made by the manager of the spots. Hence, the spots may be on private property or at the urban curb.
Happy to read your analysis. Waymo’s progress is impressive, and their operational data is invaluable. But the numbers also highlight a deeper issue: today’s vehicle width is inherently wasteful for the way urban mobility now works. Even the most efficient AV fleet still occupies far more road and curb space than is necessary to move a single person.
The 39-inch-wide Tango EV already proves what’s possible when we rethink vehicle geometry. A tandem, one-person-wide footprint dramatically reduces the road-width each vehicle consumes—whether driving, waiting, or parking. That design advantage scales directly into lower congestion, faster trip times, and a smaller curb footprint for commuter cars, highway patrol fleets, and autonomous services.
If AV networks are going to grow from thousands to hundreds of thousands of vehicles, optimizing width isn’t optional—it’s foundational. #ThinMobility https://youtu.be/8VPz38Xkgsc?si=tE23hG1C-TecM8nM