“Robotaxi” covers several products that do different work. A rider can hail a driverless Waymo in some cities. Tesla reports unsupervised rides in selected markets, while its Bay Area service still has a safety driver. Uber or Lyft may supply the booking app, the fleet operation, or both. I start with what a passenger can actually book.

This map reflects the September 2026 evidence linked below. The dated posts keep each announcement and later revision in context.

Waymo has rides to measure

Waymo reported more than half a million weekly rides across 15 U.S. cities in September 2026. Access was still expanding in the newest markets. A city on the list did not mean every resident could book a trip.

Its Detroit, Las Vegas, and San Diego expansion shows why the rollout stages matter. Naming a city, testing there, inviting riders, and opening access are separate milestones. The Swiss Re claims comparison gives the safety case evidence beyond my own comfortable ride in San Francisco.

Waymo also uses other companies’ distribution and fleet work. Uber’s Austin and Atlanta partnership combines Waymo’s driving system with Uber’s app and vehicle operations. Lyft’s Nashville arrangement put fleet work ahead of Lyft app access. The same driving system can sit inside different businesses.

Tesla needs a city-by-city read

Tesla’s Bay Area launch offered paid rides with a safety driver. The California DMV permit list placed Tesla Robotaxi LLC in testing with a driver when checked in September 2026. Its Bay Area ride service operated under a separate TCP permit using FSD (Supervised). A testing permit alone does not authorize commercial driverless rides.

Tesla’s Q2 2026 update reported expanded unsupervised operations in Austin and July launches in Miami, Orlando, and Tampa. Those are company-reported operations outside California. Booking access, supervision, and local permission need their own check in each market.

I drive a Model 3, but owning the car does not establish the ride service’s capabilities. Cybercab production cleared a manufacturing checkpoint. Paid vehicle hours and the cost of running the fleet remain the business test.

Uber and Lyft divide the work among partners

Uber’s Lucid–Nuro commitment joins a vehicle manufacturer, a driving-system supplier, and a marketplace. The conditional vehicle purchases bring exposure to idle fleet costs as well as possible control over supply. Its May Mobility deal shows another risk: a partner can miss the proposed launch date.

Lyft’s May Mobility Atlanta pilot began with onboard standby operators. Its Tensor reservations targeted a 2027 launch. Neither established a large driverless fleet simply by appearing in the partnership list. The gig-platform map follows what those commitments mean for costs and labor.

The car can finish driving and still need help

Cruise’s October 2023 crash exposed both dangerous post-impact behavior and incomplete reporting. Motional’s deployment pause exposed the distance between a technical program and a commercial service.

Waymo’s door-closing pilot puts the physical problem at a smaller scale. The trip ends, a door stays open, and someone has to return the vehicle to service. The work behind a driverless ride includes those exceptions alongside cleaning, charging, remote assistance, and recovery.

Delivery has a different handoff

DoorDash works with Serve’s sidewalk robots, develops Dot, and uses Waymo for eligible Phoenix deliveries. Each mode has different routes, payloads, and pickup requirements. Driving the order to an address is only useful if the recipient can collect it.

Amazon’s Zoox Las Vegas service takes another approach with a purpose-built passenger vehicle. Its Hayward factory supplies a possible path to scale. Factory capacity still has to become vehicles available for rides.

I follow paid rides per vehicle and the hours each car can accept bookings. Intervention frequency tells me what happens between those two numbers. A partnership earns a place in the service column when a rider can use it; a fleet earns its business case when enough paid work survives the downtime.