2026-06-18 👁️ — views

AV Fleet Operations — How Waymo Runs a Robotaxi Fleet Day-to-Day

Dispatch algorithms, charging logistics, maintenance cycles, and vehicle recovery — converting AV hardware into a profitable fleet service.

Article 68 in the Physical AI Benchmark Series — The Fleet Operations Layer

Launching a robotaxi is only half the job. The harder half — and the one that directly determines whether a robotaxi business generates returns or burns cash — is running the fleet once it is deployed. Dispatch algorithms, charging logistics, cleaning cycles, maintenance scheduling, remote monitoring, and vehicle recovery are the operational problems that convert hardware into a working service business.

Waymo has been solving these problems in commercial service since 2018 and at full driverless scale since 2022. Tesla’s Cybercab must build this entire operational layer from scratch, in multiple cities, at a pace that justifies the capital invested. The operational competency gap between these two positions is as significant as the technology gap — and it is largely invisible to investors who focus exclusively on autonomous miles driven.

Utilization rate is the key unit-economics lever in fleet operations. A vehicle that sits idle is pure cost. The operational layer — dispatch, charging, maintenance scheduling, recovery — exists to maximize the fraction of time each vehicle is earning revenue. Waymo’s stated target is 70% or higher utilization at scale (est.); achieving it requires every operational subsystem to function reliably at the city-fleet level.

Section 1 — The Dispatch Problem

Dispatching a robotaxi fleet is orders of magnitude more complex than dispatching human ride-share drivers. Human drivers self-optimize: surge pricing signals where to go, they choose their own routes, and they can exercise judgment about which rides to accept. An AV fleet has none of these properties — every positioning and routing decision must be made algorithmically.

Dispatch algorithm key variables (est.):

• Vehicle state: charge level (%), maintenance due in X miles, current location, availability status
• Demand forecasting: historical patterns by hour/day/event, real-time booking rate
• Coverage density: minimum vehicles per grid cell to ensure acceptable wait times
• Deadhead efficiency: minimize empty miles driven to reach next pickup

The deadhead optimization problem is particularly important because deadhead miles — empty driving to reach a passenger — represent pure cost with zero revenue. A fleet that is geographically misallocated relative to demand patterns will drive significant empty miles even at high booking volume. Waymo’s dispatch system uses historical demand patterns layered with real-time signals to pre-position vehicles before demand materializes, rather than reacting to trip requests after the fact.

Section 2 — Charging and Energy Management

Energy management is the operational constraint that most distinguishes a commercial AV fleet from a personal EV. A personal EV owner can decide to skip a trip to go charge. A commercial AV fleet cannot afford that flexibility — an unscheduled charge event takes a revenue-generating vehicle out of service and disrupts coverage density.

Waymo’s charging infrastructure represents a significant fixed-cost investment in each city. Proprietary charging depots must be sited, built, and maintained — with vehicle lifts, power infrastructure, and service bays for basic maintenance alongside the chargers. The mid-day trough in ride-share demand (roughly 10 AM to noon in most cities) is when Waymo rotates vehicles through the depot for charging and sensor cleaning, making efficient use of low-revenue hours.

Tesla’s Supercharger network is potentially a meaningful operational advantage for the Cybercab. Thousands of existing high-power Supercharger sites could serve as distributed charging points, reducing the capital required to build dedicated charging depots in each launch city. Whether the Supercharger network density is sufficient for commercial fleet operations — and whether the priority access logistics can be managed without disrupting consumer Supercharger availability — remains to be demonstrated.

Section 3 — Maintenance and Cleaning

A commercial AV fleet operates its vehicles at roughly 5–10 times the annual mileage of a personal car. This accelerates wear on every mechanical component and creates a maintenance operation that has no consumer-car equivalent. The addition of a sensor suite — LIDAR, cameras, radar — creates a maintenance burden that conventional fleet operators have never encountered.

Sensor cleaning is the most operationally distinctive element of AV maintenance. A dirty camera lens or LIDAR window can degrade perception performance enough to affect safe operation — this has no equivalent in conventional fleet maintenance. Waymo’s operations (est.) include automated wash stations that clean sensor housings on depot entry, with visual inspection checkpoints for residual contamination.

Sensor calibration after collision is the highest time-cost maintenance event. Even a minor low-speed collision requires a full sensor suite review and recalibration before the vehicle can return to driverless service. Calibration targets must be precisely positioned, each sensor modality calibrated in sequence, and the integrated system validated — a process estimated at 1–4 hours depending on the calibration rig and the sensors involved. For a large fleet, managing the throughput of the calibration bay becomes a meaningful scheduling challenge.

Interior cleanliness without a driver is a passenger experience problem with operational cost implications. Conventional taxi and ride-share companies rely on drivers to maintain some baseline cabin standard between rides. An AV has no such self-cleaning property. Waymo’s approach requires structured cleaning cycles — light cleaning between passengers where feasible, deeper cleaning at scheduled depot visits — adding a staffed labor cost component to what might otherwise appear to be a fully automated operation.

Section 4 — Vehicle Recovery and Incident Management

Every fleet operating at scale will experience breakdown, collision, passenger incidents, and software faults. The quality of recovery operations — how quickly a vehicle is returned to service, how well incidents are managed, how rapidly software issues are diagnosed and resolved — is a direct determinant of fleet utilization and customer experience.

Waymo operates a Remote Operations Center (ROC) that monitors the entire driverless fleet in real time. When a vehicle encounters a situation it cannot navigate — an unusually placed obstruction, an ambiguous intersection, a construction zone outside its operational design domain — a remote operator can observe the situation via onboard cameras and provide guidance. This is not remote driving in the traditional sense; the ROC is providing guidance inputs, not controlling the vehicle. But it does mean that “fully driverless” operation still requires a human monitoring infrastructure, with staffing costs that scale with fleet size.

The OTA (over-the-air) software update capability is both an operational advantage and a risk management tool. When a software fault is identified in the fleet — whether from a novel road scenario, a perception edge case, or a planning failure — Waymo can deploy a patch to the entire fleet within hours. This is fundamentally different from conventional automotive recalls, which require physical service visits. The flip side is that a software regression deployed fleet-wide can degrade the entire fleet simultaneously, making software change management a critical operational discipline.

Section 5 — What Tesla Must Build for Cybercab Operations

Tesla has mastered consumer car manufacturing at scale. Commercial fleet operations is an entirely different competency set — one that Waymo has spent six years building and that Tesla does not yet possess at any meaningful scale.

Tesla’s vertical integration is the most legitimate argument that it can close the operational gap faster than the six-year head start suggests. Tesla controls the car, the battery technology, the charging network, the software stack, and the manufacturing process. This integration eliminates many of the coordination costs that would slow a conventional automaker trying to build an AV fleet operation. The Supercharger network in particular — over 50,000 stalls globally as of 2025 — provides a charging infrastructure foundation that Waymo had to build from scratch at each new city launch.

The counter-argument is that operational competency is not primarily a technology problem. The day-to-day logistics of dispatching cleaning crews, managing parts inventory, running calibration bay throughput, and staffing remote operations centers are human-intensive operations management problems. Tesla’s experience base here is consumer car delivery and service center operations — useful preparation, but not the same as commercial fleet operations at the density and intensity required for a high-utilization robotaxi service.

The utilization gap between consumer EV ownership (roughly 4–5% utilization for a personally-owned car) and commercial fleet targets (40–70%, est.) illustrates the operational intensity required. Every percentage point of utilization improvement requires better dispatch, better maintenance scheduling, faster cleaning cycles, and quicker incident recovery. The fleet operations layer is where the robotaxi business model is won or lost.

Section 6 — Investor Signal

Fleet operations is the most under-analyzed dimension of the robotaxi investment thesis. Every major investor analysis covers miles per disengagement, geographic expansion pace, and technology partnerships. The operational infrastructure that converts an AV system into a revenue-generating fleet service receives a fraction of that attention.

Utilization rate is the unit-economics lever that matters most. At low utilization — below 40% (est.) — the revenue per vehicle per day is insufficient to cover the total cost of vehicle ownership, depot infrastructure, maintenance labor, remote operations staffing, and software amortization. At 70% or higher utilization (est.), the unit economics become attractive. Every operational improvement — faster charging, tighter maintenance scheduling, better dispatch pre-positioning — is worth meaningful dollars per vehicle per day.

Waymo’s six-year operational learning curve is a real moat. The accumulated knowledge of what breaks, how to fix it quickly, how to staff recovery teams efficiently, how to schedule depot throughput without creating bottlenecks — this institutional knowledge is not reflected in any public disclosure but represents years of expensive trial and error. Tesla must pay its own tuition to acquire it.

Tesla’s Supercharger network is a genuine operational asset. The distributed charging infrastructure reduces the CapEx required to launch Cybercab service in a new city significantly. This is the one area where Tesla enters commercial fleet operations with a demonstrable infrastructure advantage over a greenfield build.

The maintenance depot requirement creates a city-level fixed-cost structure. Neither Waymo nor Tesla can operate a robotaxi fleet in a city without a facility equipped for EV charging, sensor calibration, and vehicle maintenance. This creates a minimum-scale requirement in each city — below a certain fleet size, the fixed costs of the depot cannot be distributed across sufficient utilization hours. Fleet density in each operating geography matters more than total global fleet size for unit economics.

Section 7 — About This Series

This is article 68 in the Physical AI Benchmark Series. Previous articles have covered the ramp index, the humanoid race, unit economics, global competition, HD mapping, software and OTA, insurance and liability, consumer demand, competitive moats, Cybercab versus Model Y, safety data, Waymo Gen 6, Optimus manufacturing, scorecard snapshots, 2030 forecast scenarios, the investor framework, city expansion pipelines, Tesla FSD state approval maps, AV weather and climate constraints, the talent war, regulatory calendars, robotaxi fare pricing, humanoid deployment trackers, supply chain analysis, consumer adoption demand index, valuation and IPO analysis, the Physical AI 2026 mid-year roundup, AV unit economics cost-per-mile breakdown, the AV data flywheel comparison, AV cybersecurity attack surfaces, and the Physical AI supply chain.

This article adds the fleet operations layer: dispatch algorithms and the deadhead optimization challenge, charging and energy management (Waymo depots vs. Tesla Supercharger network), maintenance and cleaning operations (sensor calibration as the distinctive AV maintenance burden), vehicle recovery and incident management (ROC infrastructure, OTA fault response), and the operational competency gap between Waymo’s six-year commercial operations and Tesla Cybercab’s greenfield build requirement.

Note: Utilization rate estimates, maintenance intervals, charging time figures, and mean-time-to-recovery targets are labeled “(est.)” and reflect publicly available reporting, industry analysis, and analyst estimates. Actual Waymo operational parameters are not publicly disclosed. This article does not constitute investment advice.

Sources

• Waymo operational overview — Waymo safety and operations ↗
• Tesla Cybercab fleet operations — Tesla AI Day / investor presentations ↗
• EV fleet charging infrastructure — Rocky Mountain Institute ↗
• AV fleet management operations — RAND Corporation ↗

Tags