How much of a robotaxi's driving is empty?
California requires carriers to file their miles split by whether a passenger was aboard. Nobody publishes the ratio. It is the closest thing the public record holds to a unit-economics figure.
Empty share by programme
A pilot programme running mostly empty is a vehicle learning a city. A deployment programme running mostly empty is a cost problem. The two are not comparable and are never added together.
What counts as an empty mile
Carriers file their vehicle miles in three periods. Two of them describe driving with no passenger aboard — travelling to a pick-up, or repositioning between trips. The third is the passenger-carrying leg. The empty share is the first two over the total.
A month in which a carrier opens new territory reads as a spike, because vehicles spend that month repositioning. That is a maturity curve, not inefficiency.
California only. No other state requires this split.
Monthly miles — Waymo, deployment · driverless
31 months, 2023-08 to 2026-06. This is the longest run of operating data in the public record, and it belongs to a carrier that no longer operates — which is why it is the only programme whose whole life can be drawn. The filing has month resolution and nothing finer.
Ridership and occupancy
Trips, passengers and passenger-miles as filed in California — the demand side of the same record the deadhead ratio is drawn from.
A load factor near one means a robotaxi is carrying about one person. Read against the empty-share above, it is the closest the public record comes to a utilisation figure: miles are spent moving a vehicle, and only some of those miles are spent moving a passenger. Months where a carrier filed miles but withheld trips are left out rather than counted as zero.