Waymo, Zoox, and Tesla: Different Approaches and Operational Implications of Self‑Driving Cars

This Week’s Focus: Three Roads to Robotaxis

Zoox is expanding its robotaxi service from Las Vegas to San Francisco, while Waymo, Alphabet’s AV unit, already runs driverless services in Phoenix, LA, SF, and Austin, and is testing airport routes. Tesla, meanwhile, offers paid rides in a small Austin zone using modified Model Ys. Three firms, three models—purpose-built fleets, multi-sensor retrofits, and camera-only cars. This week, we examine their sensor strategies, vehicle designs, economics, and operational challenges, and ask: which approach is best positioned to win the race for autonomous ride-hailing?

Last week I took a Zoox ride in San Francisco. Long‑time readers may recall that I was once a self‑driving skeptic, but I’ve since become a convert. The experience blew me away. The vehicle looks futuristic, and whether you’re facing forward into traffic or sitting backward with other passengers, the ride is unexpectedly comfortable. Uneventful even…in the best possible way!

Zoox already runs a public robotaxi service on the Las Vegas Strip and is preparing to launch another one in San Francisco. For now, only Zoox employees can request a ride in SF—I was fortunate to be invited by Steven Lord, a former student and long-time Zoox employee, to join one.

But Zoox is not alone.

Waymo, Alphabet’s self‑driving subsidiary, operates driverless robotaxis in Phoenix, Los Angeles, San Francisco, and Austin, among other cities. It recently secured a permit to begin testing rides to San Francisco International Airport, expanding on its existing airport operations at Phoenix’s Sky Harbor and ongoing testing at San Jose Mineta.

Tesla is also offering paid robotaxi rides in a limited zone of Austin, TX, using Model Y vehicles modified to include safety monitors.

All this got me thinking what interesting times we live in: three AV ride-hailing firms using three distinct models. These approaches set the stage for a comparison of their sensor strategies, vehicle designs, economic models, ownership structures, and operational challenges.

In today’s article we explore the differences between these models, their advantages and their limitations, and through academic research, we try to predict which one is more likely to prevail.

Let’s take a test drive.

Sensor Strategies: LiDAR and Redundancy Versus Camera‑Centric Designs

A main difference between the three firms is their approach to sensor strategies.

Waymo: multi‑sensor redundancy: Waymo’s 6th‑generation Waymo Driver hardware uses 13 cameras, 4 lidars, 6 radars, and external audio receivers to create an overlapping 360‑degree field of view. The vehicle detects objects up to 500 meters away even in darkness or poor weather. Redundant actuators provide backup steering and braking so that a single hardware failure will not cause loss of control. The system is modular: sensors are positioned at the four corners and on the roof, with units that can be cleaned and heated for cold or dusty conditions. A high‑resolution map and continuous data logging feed AI algorithms that constantly learn from real‑world and simulated miles. As MotorTrend notes, Waymo’s robotaxis use “a multitude of radar, lidar and optical sensors” to give AI real‑time perception for moment‑by‑moment driving decisions.

Zoox: purpose‑built architecture with sensor overlap: Zoox designed its robotaxi from scratch, equipping it with a sensor architecture that combines cameras, radars, and lidars. Sensors are mounted on each corner and the roof to create a 360° overlapping field of view, capable of detecting objects more than 150 meters away in all directions, including around corners. The vehicle has bidirectional driving (no defined front or back), four‑wheel steering, and a fully electric powertrain designed without a single point of failure. The battery supports up to 16 hours of operation and the system includes backup components (e.g., redundant power and steering) to ensure safety. Exterior lighting cues signal braking and turning to nearby road users.

So far, the similarities are clear.

Tesla: camera‑only “pure vision”: Tesla’s autonomous strategy diverges sharply from its rivals. The company eliminated radar and LiDAR in favor of a camera-only system called “Tesla Vision.” Reuters notes that Tesla is the only major autonomous driving company to abandon radar and LiDAR, relying entirely on cameras for sensing.

Elon Musk argues this approach will be safer and cheaper. In its Austin robotaxi pilot, Tesla deployed a small fleet of Model Y vehicles, equipped with safety monitors in the front seats, restricted to a limited geographic area. According to Musk, the camera‑only system and internally developed AI chips underpin the autonomy. While a camera‑centric design may reduce sensor costs, it lacks the redundancy and all‑weather reliability of LiDAR‑based systems, raising questions about robustness under adverse conditions.

One of Elon Musk’s arguments is that cars shouldn’t have radars, since people don’t have radars. But last time I checked, people don’t have cameras…or wheels either. I understand the cost aspect though.

In practice, these sensor strategies have produced very different safety records.

Waymo’s multi‑sensor redundancy translates into real‑world safety gains. According to the company’s latest Safety Impact dashboard, its robotaxis have logged over 100 million real‑world miles and recorded 91% fewer severe‑injury crashes, 79% fewer airbag‑deployments, and 80% fewer injury‑causing crashes than typical human drivers across its service areas. Collisions involving vulnerable road users are more rare, with 92% fewer pedestrian‑injury crashes, 78% fewer cyclist‑injury crashes, and 89% fewer motorcycle‑injury crashes.

Zoox’s record is still emerging. The company boasts a purpose‑built vehicle with redundant systems, but has already reported two minor incidents—an April 8 crash (no injuries) with a passenger car in Las Vegas that triggered a recall of 270 vehicles and a brief service pause, and a May 8 incident in San Francisco where a scooter rider sustained minor injuries after crashing into a parked Zoox robotaxi; Zoox temporarily halted operations and issued another software update.

Tesla’s camera‑only robotaxi program has the smallest sample size and the worst early results: a federal filing revealed that the Austin pilot fleet suffered three crashes within its first 7,000 miles, including two rear‑end collisions and a crash into a stationary object that caused minor injuries and required towing. Videos of Teslas speeding and ignoring traffic laws also triggered a safety investigation by the NHTSA.

Vehicle Design: Retrofit Versus Purpose‑Built Robotaxis

Another key area of divergence of these firms is vehicle design and manufacturing.

Waymo’s retrofitted luxury EVs: Waymo integrates its driver technology into the Jaguar I‑Pace electric crossovers, producing a comfortable ride (in my experience and anyone I spoke with) within a conventional cabin. By retrofitting an OEM vehicle rather than building its own, Waymo leverages established automakers’ safety certifications, crashworthiness, and supply chains. However, this strategy requires partnerships and may constrain vehicle layout, since the cars still have steering wheels and pedals (though disengaged), limiting cabin space.

Zoox’s toaster‑shaped robotaxi: Zoox has engineered a purpose‑built robotaxi with no driver seat, steering wheel or pedals. The vehicle is symmetrical: it can drive equally well in either direction and seats four passengers facing each other, with subway‑style sliding doors. The design maximizes interior space and integrates sensors, computing hardware and battery packs into a compact package. Zoox emphasises seat belts, wireless charging, touchscreens, and climate control.

Tesla’s interim and future vehicles: In 2025 Tesla launched an invite‑only robotaxi pilot in Austin using modified Model Y SUVs. These vehicles retain steering wheels and pedals, with a safety monitor in the front seat to take control if needed. Tesla has announced plans for a purpose‑built “CyberCab” with no steering wheel or pedals, targeted for production in 2026. Musk has also hinted at a “Robovan” for ride‑sharing and goods delivery, and envisions owners renting out their personal Teslas as robotaxis when idle. Tesla’s approach thus spans retrofits (existing Model Y) and purpose‑built vehicles, balancing near‑term deployment with longer‑term ambitions.

The main implication is cost. By building on existing vehicle platforms and only using cameras, Tesla will dominate in terms of cost. The main tradeoff comes in ride comfort and safety.

Ownership Structures and Their Cost Implications

Waymo—Fully owned fleet with partnerships: Waymo operates a fully owned fleet of robotaxis and controls the entire stack—from sensor hardware and software to fleet maintenance and operations. Alphabet’s financial strength enables heavy investment in R&D and long test cycles. Waymo has also partnered with Uber to offer rides through the Uber app in several cities, increasing utilization without ceding control of its technology. The company has signaled that it may eventually offer its autonomous driver technology to automakers for personal vehicles.

Owning the fleet allows Waymo to maintain high safety standards and collect large datasets, but it also means high capital expenditure and slower scaling compared to asset‑light models.

Zoox—Amazon‑owned robotaxi service: Zoox is a wholly owned Amazon subsidiary with a strategy resembling a vertically integrated mobility service: it designs and builds its own vehicles, operates the fleet, and manages the app. With no steering wheel or driver seat, Zoox’s vehicles are not designed for individual ownership but purely for robotaxi service.

Amazon can integrate Zoox with its broader logistics and retail ecosystem (e.g., using robotaxis for Prime deliveries during off‑peak hours). Full ownership allows Zoox to tailor every component for autonomy, but requires large upfront investment in manufacturing facilities and service depots.

Tesla—Hybrid model and peer‑to‑peer platform: Tesla currently sells vehicles to consumers and monetizes the Full Self‑Driving (FSD) software as a pricey option.

-Interim retrofit fleet: The Austin pilot uses Tesla‑owned Model Y vehicles with FSD and safety monitors. Tesla controls dispatching and collects fares.

-Future CyberCab fleet: Tesla plans to build a purpose‑built robotaxi (CyberCab), which could potentially be owned by Tesla or sold to fleet partners.

-Peer‑to‑peer network: In the scenario where privately-owned Teslas are deployed as robotaxis when idle, Tesla would take a percentage of fares while owners bear the cost of vehicle purchase and maintenance. Such a model could accelerate fleet growth and reduce capital requirements, but it also raises regulatory and insurance challenges.

You can already see some of these implications in how these services are priced:

Waymo positions its robotaxis as a premium product: a San‑Francisco price‑comparison study of ~90,000 ride quotes found the average Waymo trip cost $20.43—$5.99 more than Lyft and $4.85 more than Uber, making Waymo rides 41% and 31% more expensive, respectively. Even though riders praised the cleaner vehicles and smoother driving, only about 43% were willing to pay extra. The high fares reflect the cost of multi‑sensor hardware, mapping, cleaning, remote assistance, and a limited supply of vehicles.

Zoox, by contrast, launched its Las Vegas service with free rides while awaiting regulatory approval, operating a fleet of about 50 purpose‑built robotaxis on a handful of Strip destinations. Amazon’s deep pockets allow Zoox to subsidize trips as it builds a user base, but long‑term profitability will depend on manufacturing cost, battery life and utilization.

Tesla takes a different route: the Austin pilot charged a flat fee of $4.20 per ride and used roughly ten Model Y vehicles within a small geofenced area. This teaser price signals a low‑cost strategy made possible by using existing cars and fewer sensors.

Operational Challenges: Maintenance, Charging, and Depot Logistics

Deploying autonomous fleets at scale requires far more than perception and planning. Robotaxi operators must keep vehicles cleaned, charged, maintained, and connected—without a human driver to plug in or wipe down a sensor.

Let’s look at how each company handles these “hidden” jobs and the trade‑offs built into their approaches.

Waymo: Sensor Care, Charging, and Outsourcing Depot Operations

Sensor maintenance. Because Waymo vehicles operate for long stretches without manual intervention, the company integrates preventive cleaning systems. Sensors can be heated and cleaned to remove dust and ice, and maintenance routines adapt to colder climates. Safety margins are built into its sensor suite to keep cameras, LiDARs, and radars reliable in heat, fog, rain, and hail. These protections add cost and complexity, and cleaning hardware must withstand bugs, dust, and ice.

Charging and data. Operating an all‑electric robotaxi fleet requires depots with high‑power DC chargers and high‑speed data links. Analysts note that apart from parking, AV depots must provide EV charging and fast internet to upload logs. Waymo’s primary San Francisco depot installed 38 DC fast chargers of around 60 kW each, implying roughly 2.4 MW of site power. Building such infrastructure involves expensive real‑estate, permitting, and utility upgrades. Depots should be close to demand to minimize deadheading (non‑revenue mileage), but these locations often face community opposition.

Fleet management partnerships. To scale more quickly, Waymo has begun outsourcing depot operations. In Austin and Atlanta the company’s expanded partnership with Uber gives Uber responsibility for fleet management—cleaning, repairs, and general depot operations—while Waymo focuses on the autonomous driver and roadside assistance. In Phoenix and Miami, Waymo is partnering with Moove to manage fleet operations and charging infrastructure. Outsourcing accelerates expansion but creates dependence on partners and clear lines of responsibility.

Zoox: Mission‑readiness Teams and Depot Build‑out

For Zoox’s purpose‑built robotaxi, everyday tasks like wiping sensors or plugging in chargers must be done by staff. The company operates a depot and office in Las Vegas, and is expanding it by 190,000 ft² to house vehicles and support crews. Most of Zoox’s new hires there join a mission‑readiness team responsible for starting and shutting down vehicles, cleaning sensors and interiors, swapping components and charging batteries. Unlike Waymo, Zoox is vertically integrated, funding its own real‑estate, infrastructure, and maintenance. Remote human assistance is available for unusual situations, but the company must be able to intervene quickly when a vehicle encounters a problem.

Tesla: Distributed Charging and Owner Involvement

Tesla’s operational model differs from Waymo and Zoox because most of its vehicles are privately owned. In the Austin robotaxi pilot, the safety monitor installed in the front seat plugs in the car, wipes cameras, and ensures the vehicle is charged between rides—tasks that will need to be automated or staffed when monitors are removed. For future CyberCab deployments, Tesla plans to leverage its Supercharger network rather than build large depots—owners or Tesla staff can use existing infrastructure. Maintenance and repairs are handled at Tesla’s service centers, and data from Full Self‑Driving is uploaded over the vehicle’s cellular connection. Without a central depot, Tesla can scale more quickly but has less control over cleaning and upkeep—making consistent service quality across owner‑operated cars challenging.

These operational challenges show that autonomy is not just a software problem. Even with a safe driving stack, scaling robotaxi fleets requires expensive real estate, high‑power charging, high‑bandwidth connectivity and a workforce to perform cleaning and maintenance. Different ownership models shape how firms choose to tackle these hurdles: Waymo and Zoox invest in depot infrastructure, while Tesla relies on distributed networks and owner involvement. Addressing these pragmatic issues will be as critical as refining perception and planning in bringing self‑driving cars to the masses.

Passenger Experience Across Waymo, Zoox, and Tesla

At the end of the day, the question is whether enough people will adopt robotaxis. Early data show that early passengers are using them frequently, but the jury is still out. For mass adoption, passenger experience will be critical.

Waymo: MotorTrend’s 2025 test of Waymo One notes that once passengers settle into the Jaguar I‑Pace EV crossovers, the novelty of a driverless car gives way to relaxation. The app is simple to use, and passengers never need to negotiate with a human driver about music or temperature (or why it’s important to be dropped off at my actual address and not across the street). Touchscreens allow climate and music control or access to support, while interior cameras monitor safety. The drive is smooth and vehicles demonstrate situational awareness (there are on-screen visuals of the objects being detected).

Zoox: Amazon’s preview highlights the spacious, communal cabin—no steering wheel, no pedals, just face‑to‑face bench seating. Passengers buckle up with built-in seat belts, enjoy cupholders, wireless charging for two phones, USB ports for laptops or tablets, and individual climate controls. Passengers can set the music via a central screen and the ride begins with a swipe of the screen. The vehicle travels at a comfortable speed, while obeying traffic laws, and riders quickly forget that there’s no driver. A horseshoe‑shaped airbag system surrounds each seat for added safety.

Tesla: First‑person accounts of Tesla’s Austin pilot describe robotaxis as brand-new Model Y SUVs with white interiors, panoramic sunroofs, front light bars, and pristine cabins that are frequently cleaned. During beta testing, a human safety monitor sits up front. Passengers request the ride through the app, and the car waits up to 15 minutes at a safe pick‑up location. Inside, there are two screens: a large front display showing the car’s route and traffic, and a smaller rear screen that lets riders start the ride, access their phone’s apps, or add music. Vehicles include a “Support” button that connects riders to a human operator, and a “Pull Over” button that can halt the car in an emergency. Reviewers found the ride smooth and calm, with the camera‑only system showing traffic and obstacles on the screen. Passengers appreciated the ability to customize seat positioning via their app and noted that preferences were stored for subsequent rides.

Public perception: Surveys suggest that public apprehension remains a significant hurdle. A Sherwood News report on consumer sentiment found that about 70% of Americans wouldn’t ride in a robotaxi, and roughly half said Tesla’s Full Self‑Driving technology should be illegal. The same survey reported that nearly 90% would blame Tesla if a self‑driving car caused a fatal crash, and 71% favored regulations requiring both cameras and lidar instead of camera‑only systems. These attitudes reflect anxiety after high‑profile crashes and underscore the importance of transparent safety metrics and redundant sensing.

Passenger comfort: Self‑driving vehicles free riders to read, work, or relax, but that lack of visual engagement can exacerbate motion sickness. Research from the University of Michigan shows that one in three adults gets carsick—a problem that could worsen as drivers become passengers in autonomous cars. Engineers there are developing PREACT, a system that uses sensors to predict vehicle motions and vibrates or tilts the seat to cue passengers ahead of turns and stops. Their tests indicate the technology can reduce motion‑sickness symptoms by nearly half. Such human‑factor innovations may be crucial for widespread adoption.

Ride smoothness: Vehicle behavior varies by platform. MotorTrend found Waymo’s Jaguar‑based robotaxis to drive cautiously and smoothly, with a balanced mix of defensive and assertive maneuvers. My experience with Zoox is that the vehicle travels at a comfortable speed, and as a passenger I quickly forgot there was no driver; its flat floor and four‑wheel steering help reduce abrupt motions. I easily get motion sickness when I ride in the backseat of a Tesla. The Zoox ride felt very comfortable. Tesla’s Austin rides were described as smooth and calm, though some passengers reported nausea—likely due to the camera‑only system’s sudden lane changes or braking and the Model Y’s relatively firm suspension. Future robotaxis may mitigate these issues by tuning suspension for ride comfort and by providing visualizations of upcoming maneuvers on in‑car screens, helping passengers anticipate motion.

Who will Prevail?

At this point it’s clear that no single advantage guarantees overall victory.

If hardware costs fall sharply—for example, if solid‑state lidar becomes cheap and robust—fully owned robotaxi fleets like Waymo’s and Zoox’s could deliver superior safety at prices closer to ride‑hail services, making them more competitive.

Conversely, if camera‑only autonomy proves sufficiently safe, Tesla’s distributed model could dominate through lower costs and rapid scaling. Should regulators mandate multi‑sensor redundancy for Level 5 autonomy, Waymo or Zoox would be better positioned, whereas laxer rules could favor Tesla’s cost‑driven approach.

In the most likely scenario, analysts expect a mixed market: premium, high‑safety robotaxi services in dense urban cores and lower‑cost, owner‑operated autonomous rides in suburban and rural areas.

At this point it’s clear that In the robotaxi race, the finish line isn’t a line on a map but a seatbelt click in millions of cars; whoever convinces the public that letting go of the steering wheel is safer, cheaper, and more comfortable will shape the future—regardless of the number of sensors they carry. And as my readers know, I’m among those who trust this technology more than I trust most Uber drivers (or myself).