AV Depot Operations: 24/7 Cleaning, Charging & Dispatch

A Day at a 100-Vehicle AV Depot

Walk into a 100-vehicle Waymo, Zoox, or Cruise depot at 3 a.m. and the building is busy. Twelve vehicles are in active wash-bay rotation, twenty-eight are on chargers, eighteen are inspecting and rotating in and out of dispatch staging, and forty-two are on revenue service in the operating area being summoned from the dispatch queue. The lights are on but no drivers are present anywhere on the floor. Two dispatchers sit in a glass-walled operations room watching a multi-screen view of the depot floor, the chargers' real-time power draw, the cleaning bay's throughput, and a per-vehicle dashboard showing each AV's state in the rotation. A cleaning supervisor walks the floor every 90 minutes spot-checking interior conditions. A facility technician is dispatched to a charger that has thrown a recoverable fault. A teleoperator at a remote operations center 800 miles away has just taken control of a vehicle approaching the depot from an active construction zone on a side street.

The rhythm is steady and continuous because robotaxi demand does not stop. A modern AV fleet in Phoenix, San Francisco, or Austin runs heavy passenger demand from 6 a.m. to 11 p.m. and meaningful airport and late-night demand from 11 p.m. to 5 a.m. The depot's job is to keep vehicles charged, clean, and dispatchable around the clock, with peak vehicle-cycle throughput between 1 a.m. and 5 a.m. when the largest share of the fleet is rotating through depot maintenance windows. A typical fleet vehicle returns to the depot 2 to 4 times per day for charging and 1 to 2 of those returns will include a full cleaning cycle. The math: 100 vehicles, 3.5 average cycles per day, equals 350 vehicle-cycles per day at a steady-state depot — close to the throughput design point we use when sizing facility infrastructure.

What is most striking inside an AV depot is not the technology but how few humans are present. The 4 to 8 people per shift at a 100-vehicle depot are a small fraction of the headcount at a comparable traditional taxi or ride-share garage, which might run 25 to 40 people across the same shifts to handle dispatch, driver pickup and drop-off, fueling, light maintenance, and customer service. The cost differential is the central economic argument for the AV business model — and the central operational challenge for the parking operator, which must run a high-throughput, high-uptime facility on much thinner per-cycle margins than a traditional parking facility.

Robotic Cleaning Bay Throughput

The cleaning bay is the single biggest physical bottleneck in an AV depot and the place where the gap between a generic garage retrofit and a purpose-built AV operations facility shows up most starkly. A modern robotic exterior wash bay from Tommy Car Wash Systems, RYKO Solutions, or Sonny's Enterprises throws a passenger vehicle through the entire wash-and-dry cycle in 3 to 5 minutes when configured for AV use, compared to 7 to 9 minutes in a generic commercial wash bay. The throughput gain comes from omitting the customer-service overhead — no payment transaction, no driver instruction, no upsell — and from optimizing for the geometry of a known AV model rather than the worst-case driver-operated SUV. A 5-minute wash bay running continuously delivers 12 wash cycles per hour per bay, or 288 per day at steady state. Most 100-vehicle depots run two parallel wash bays, sized for 575 wash cycles per day capacity against a typical demand of 350 to 450.

Interior cleaning is the variable that requires more attention. Most depots inspect interior condition via cabin cameras after each passenger drop-off — capturing high-resolution images of seat surfaces, floor mats, cup holders, and door pockets — and route the vehicle directly to the next dispatch if no anomaly is detected. The camera-rated cleaning score is calculated against a baseline reference image, and any deviation above a configured threshold flags the vehicle for a human cleaning intervention at the next depot return. Roughly 18 to 28 percent of returns require a human cleaning touch in current operations. That fraction is projected to decline as cleaning robots improve and as passenger behavior adjusts to AV cabin norms, but it will not reach zero in the foreseeable future.

The human cleaning tech's role at an AV depot is fundamentally different from a traditional taxi or rental car cleaning role. A traditional cleaner spends 8 to 12 minutes per vehicle handling routine vacuuming, wipedown, and trash removal across every vehicle. An AV depot cleaner is dispatched only on flagged vehicles, handles a focused 3 to 5 minute targeted intervention, and moves to the next flagged vehicle. The cleaning tech's productivity is therefore much higher — typically 35 to 50 flagged interventions per 8-hour shift compared to 20 to 30 full cleanings in a traditional facility. The labor cost per vehicle-cycle is correspondingly lower, which is part of why AV depot economics work. Our tech-only management page describes how the camera-and-dispatch model carries into other parking asset types.

Charging-While-Cleaning Sequencing

The charging stall layout co-locates with the wash bay output so a vehicle exiting the wash can be auto-parked directly into a stall without a separate transit maneuver. The charger then begins energizing within 30 seconds of vehicle arrival, either via an automated cable-connect arm (deployed at Waymo's San Francisco depot and Zoox's Foster City depot) or via wireless inductive pads (deployed in pilot quantities at Cruise's Austin depot before Cruise's 2024 operational pause). Both technologies have advantages: the automated cable-connect arm supports any standard CCS1 or NACS port and delivers 90 kW to 180 kW charging speeds; wireless inductive pads avoid the cable failure mode entirely but cap charging speed at 75 kW today with a road-map to 200 kW by 2028.

The charging stall mix at an AV depot reflects the fleet's energy budget rather than the destination-versus-corridor framing that applies to retail charging. Most depots run an 70/30 mix of Level 2 to DCFC: Level 2 stalls (12 to 19 kW) handle the long-dwell mid-day rotations when vehicles can spend 60 to 90 minutes off-route; DCFC stalls (90 to 180 kW) handle the short-dwell late-night rotations when vehicles must be back on-route within 20 minutes to meet morning peak demand. The dispatch algorithm assigns each returning vehicle to a stall type based on its state-of-charge, the time until its next predicted dispatch, and current charger availability. Optimization at this layer is where the AV operator earns most of its margin — squeezing 5 percent more cycles out of the same facility through better stall-assignment heuristics is worth substantial annual revenue.

Interior cleaning happens during the charging window. The cabin cameras capture inspection images at the moment of vehicle arrival. If the score is above the cleaning threshold, the vehicle is queued for direct dispatch as soon as charging completes. If the score is below the threshold, a cleaning tech is dispatched to the stall during the charging window so the cleaning happens in parallel with charging rather than serially. The end-to-end sequence — wash, park, charge, inspect, clean if needed, dispatch — averages 38 to 52 minutes per cycle at well-optimized depots, compared to 75 to 110 minutes at depots running serial rather than parallel sequencing. The 30-minute differential per cycle, multiplied by 350 cycles per day, is the difference between a depot that can support 100 vehicles and one that can only support 70.

Teleoperator Handoff & The Comms Stack

Teleoperators are remote human operators who can take supervisory control of an AV when the vehicle's autonomy stack encounters a situation it cannot resolve independently. Teleoperator support is most common during on-road exceptions — construction zones, emergency vehicle interactions, ambiguous gestures from pedestrians or cyclists — but a meaningful fraction of teleoperator interventions happen at the depot: arrivals during severe weather, exception handling in the cleaning bay (a passenger-left-behind item triggers a teleoperator review of the vehicle's interior), and dispatch readiness checks when the autonomy stack flags an internal anomaly. A 100-vehicle depot typically sees 8 to 14 teleoperator handoffs per day, with the trend declining over time as the autonomy stack matures.

The depot's communications infrastructure must support teleoperator handoff to the same standard as the AV operator's on-road comms — sub-10-millisecond latency, sub-1-second packet loss tolerance, redundant connectivity. Practically, this means the depot must have a fiber-optic backhaul of at least 1 Gbps with a second-carrier 500 Mbps backup, on-site WiFi-6E or 5G private cellular with sub-millisecond local-area latency, and a power-redundant networking stack with battery and generator backup. The parking operator is responsible for maintaining this stack to a 99.95 percent uptime target, which is more stringent than even the charger uptime SLA. Failure of the comms stack does not just inconvenience teleoperators; it can require pausing depot operations entirely until backup connectivity is established.

The teleoperator workforce is small and centralized. Waymo's teleoperator center in Phoenix supports the entire Waymo fleet across the U.S.; Zoox's center in Foster City supports the Zoox fleet in California and Nevada. The teleoperator-to-vehicle ratio is typically 1 to 30 at current automation levels, projected to climb to 1 to 100 by 2030 as the autonomy stack matures and fewer interventions are required. From the depot operator's perspective, the teleoperator function is a logical service consumed through the AV operator's comms stack rather than a physical service hosted at the depot — but the depot must deliver the comms reliability that makes the teleoperator function viable. Our robotaxi curb design page covers the on-road counterpart to the depot comms architecture.

How a Parking Operator Earns Its Fee in the AV Stack

The parking operator's economic value at an AV depot is not the parking — there are no human drivers paying for stalls — but the operating service stack: facility uptime, charger uptime, cleaning bay throughput, comms reliability, ground-truth fiducial maintenance, security monitoring, and KPI reporting. The AV operator pays a per-vehicle-cycle fee that bundles all of these services together, typically $4 to $9 per cycle depending on contract scope and SLA tier. At 350 cycles per day, that is $1,400 to $3,150 per day of gross revenue — $510,000 to $1,150,000 per year per 100-vehicle depot. The operator's all-in cost to deliver this service stack is typically $310,000 to $560,000 per year per depot, leaving $200,000 to $590,000 per depot in operating margin to the parking operator.

The operator's competitive advantage relative to the AV operator running the depot in-house is operational specialization. A parking operator running 10 AV depots can share dispatch supervision, facility maintenance trucks, cleaning supply procurement, fiber-redundancy contracts, and charger spare parts inventory across the portfolio — driving per-cycle cost down by 25 to 40 percent relative to an in-house AV operator running a single depot. The AV operator gets out of the business of running facility operations and focuses on its core competency of dispatch algorithm, autonomy stack, and customer experience. The parking operator earns a margin for delivering exactly the operating discipline that traditional commercial parking has always sold.

The contract structure that makes this work is performance-based with hard SLA thresholds. The parking operator earns the full per-cycle fee when all four primary KPIs are met (vehicle-cycle throughput, facility availability, charger uptime, cleaning quality score) and pays a graduated SLA credit for each KPI missed. Most contracts cap the SLA credit at 18 to 24 percent of monthly revenue to protect the operator from a catastrophic month — but the structure ensures alignment between the operator and the AV operator on the metrics that actually matter. The contracts are typically 3 to 7 years with extension options, which is long enough to amortize the depot-specific operations setup but short enough to keep the operator honest on performance. Our robotaxi depot management page walks through the broader robotaxi depot operating model.