Urban Autonomy Lab
Chapter 4 · EV115

Peer-reviewed CPUC study — ~46% of Waymo California miles run empty (deadheading)

Date collected: 8 June 2026
Summary

Abdelhalim (MIT/Aspen), *Findings* May 2026; 86.3M Waymo California VMT, 13.8M trips Aug 2023–Dec 2025 — only 53.6% passenger-onboard, ~46% deadheading; plateaued 55–57% since mid-2025, suggesting structural, not ramp-up, inefficiency

Headline Finding

A peer-reviewed study published in Findings (Transport Findings, 18 May 2026; DOI 10.32866/001c.161870) analysed 86.3 million vehicle miles across 13.79 million Waymo trips reported to the California Public Utilities Commission (Aug 2023–Dec 2025) and found that only ~53.6% of total miles carried a passenger — i.e. roughly 46% of all Waymo VMT in California is empty (deadheading). The figure has plateaued at a 55–57% passenger share since mid-2025, meaning deadheading is structural, not a transient ramp-up artefact. Picked up widely this week (Planetizen, Streetsblog USA 8 Jun, Technology.org 4 Jun).

Detail

  • Author / provenance: Awad Abdelhalim (MIT; supported by an Aspen Institute Science & Technology Policy Fellowship). Submitted 27 Apr 2026, accepted 7 May 2026, published 18 May 2026. Data source is the monthly aggregated records (trips, passengers, VMT) Waymo is required to file quarterly with the CPUC — the unredacted portion of otherwise heavily redacted reports.
  • Method: VMT disaggregated into three periods — (1) repositioning/awaiting requests, (2) en route to a confirmed pick-up, (3) passenger pick-up to drop-off. Deadheading = periods 1 + 2.
  • Key numbers:
    • Total trips Aug 2023–Dec 2025: 13,790,147; total VMT: 86,269,177; passengers carried: 19,317,109.
    • Average monthly growth ~15.2% across all three metrics (peaking ~50% in mid-2024 on LA launch + SF waitlist removal).
    • Passenger-onboard share started at just 36% in the first month of commercial service, improved as the fleet scaled, then plateaued at 55–57% since mid-2025. Overall period average: 53.6% passenger-onboard.
    • Per-trip deadheading fell from ~5.1 miles (Aug 2023) to ~2.8 miles (end 2025); inter-trip idle wait fell from 28.6 min to ~18 min — efficiency gains showed up per-trip but did not move the aggregate empty-mile ratio.
    • ~1.4 passengers per trip throughout — a low propensity for shared/pooled rides.
  • Benchmark: This roughly matches human ride-hailing (Uber/Lyft deadheading estimated >40% of VMT; Henao et al. 2019), directly undercutting the claim that autonomy reduces empty-mile inefficiency. The study cites Schaller (2021): AVs are unlikely to change auto users’ resistance to shared services.
  • A temporary trips/VMT dip in June 2025 reflects the LA service suspension during civil unrest (cross-references the legitimacy/operational-suspension theme).

Governance Implication

This is the cleanest peer-reviewed quantification yet of the Demand failure mode: robotaxi growth adds VMT roughly in proportion to ridership plus a persistent ~46% empty-mile overhead, with negligible pooling (1.4 pax/trip). For city governance the implication is that fleet scale translates almost one-for-one into additional road and kerb occupancy — empty cruising is two-thirds of the empty miles, i.e. vehicles roaming the network while waiting for a fare. This is the empirical anchor for instruments the paper already advocates: per-mile or empty-mile (deadhead) pricing, occupancy-linked fees, kerb/PUDO charging, and pooling mandates. It also strengthens the case that operator-reported data to a public regulator (the CPUC model) is itself a governance instrument — the only reason this analysis was possible is that California compels the disclosure; most jurisdictions do not. Directly relevant to the Demand discussion in Ch.5 and the congestion/value-at-risk framing in Ch.4.

Sources

Thread