The Rise of Robotaxis: How Self-Driving Taxis Are Transforming Urban Transport

Self-driving taxis are steadily evolving from experimental technology into a full-fledged mode of urban transportation. These vehicles can already build routes on their own, recognize cars and pedestrians, stop at traffic lights, change lanes, and deliver passengers without constant human intervention.

Robotaxis utilize cameras, radar, LiDAR, satellite navigation, and artificial intelligence algorithms that continuously assess the road environment. Different companies take different approaches to autonomy: Waymo relies on a complex array of sensors and detailed maps, while Tesla aims for high autonomy primarily through cameras and software.

What Is a Robotaxi and How Does a Driverless Taxi Work?

A robotaxi is a vehicle designed to carry passengers with minimal or no driver involvement. Its function is similar to a regular taxi: the passenger sets a pick-up and drop-off location, and the car completes the trip autonomously.

The main difference is that control is handled by an onboard system, not a human. The car gathers environmental data from sensors, determines its road position, analyzes other vehicles, and plans its next actions.

Robotaxi vs. Regular Car with Autopilot

The term autopilot is often used for any automatic driving system, but their capabilities vary greatly. In many production cars, the driver must still monitor the road and be ready to intervene at any time.

Robotaxis are created to independently complete an entire trip within the area and conditions for which they're designed. The car starts driving, chooses lanes, turns, reacts to traffic lights and obstacles, and stops at the destination-all on its own.

However, just because a car can drive without a driver doesn't mean it works equally well everywhere. Most modern systems are limited to specific cities, areas, weather, or road types-a restriction known as the Operational Design Domain (ODD).

Taking a Ride in a Driverless Taxi

For passengers, using a self-driving taxi is much like ordering a regular ride. Through an app, you select a pickup and destination, and the system assigns the nearest available vehicle.

When the robotaxi arrives, passengers confirm the ride and get in. The cabin usually has an interface to start the ride, adjust route options, contact support, or stop the car in an emergency.

During the trip, the onboard computer constantly processes sensor data-tracking nearby vehicles, cyclists, pedestrians, lane markings, traffic lights, and road signs-to select a safe path.

The route is not a pre-recorded sequence; the robotaxi constantly adjusts its decisions in real time, yielding to pedestrians, slowing for parked cars, changing lanes due to roadwork, or stopping if it deems the route unsafe.

Levels of Autonomy in Robotaxis

The capabilities of autonomous cars are usually described using the SAE autonomy scale from Level 0 to Level 5:

• Levels 1 and 2: The vehicle can perform specific actions (like lane keeping or distance control), but the driver is still responsible.
• Level 3: The car can drive itself under certain conditions, but a human must be ready to take over.
• Level 4: The car can make trips without a driver, but only within well-defined conditions-such as a specific city or road type.
• Level 5: Full autonomy anywhere a human driver could go, with no location or control limitations. This remains a future goal.

How a Robotaxi Sees the Road and Makes Decisions

To drive safely in a city, a self-driving taxi needs more than just a route. It must constantly understand what's happening around it-where nearby vehicles are, who's about to cross the road, the current traffic light, and whether the lane ahead is clear.

Robotaxis combine data from multiple sources. Sensors create a digital map of the environment, while software identifies objects, predicts their movement, and chooses a safe route.

Cameras, LiDAR, Radar, and Ultrasonic Sensors

Cameras work much like human vision, recognizing lane markings, traffic signals, road signs, vehicles, and pedestrians. Multiple cameras with various angles cover the car's surroundings.

LiDAR uses laser pulses to measure distances to objects, creating a detailed 3D point cloud for understanding the shapes and locations of obstacles. This is especially useful for identifying curbs, cars, poles, and other objects. LiDAR cannot alone determine signal colors or read signs, so it complements rather than replaces cameras.

Learn more about how LiDAR works in our dedicated article on LiDAR technology in smartphones and vehicles.

Radar uses radio waves to measure object distance and relative speed, working well in fog, rain, or poor lighting.

Ultrasonic sensors are used for short distances, such as parking and maneuvering near obstacles. The exact sensor suite depends on the vehicle's architecture-some manufacturers use all types, others minimize sensor count.

How Machine Vision Recognizes Cars, People, and Road Signs

Raw camera images don't inherently inform the car about the road situation-they must be processed into objects the system can understand.

Machine vision algorithms identify the boundaries of vehicles, pedestrians, cyclists, road signs, and traffic lights. The system can simultaneously track dozens of road users and assess their position relative to the car.

Importantly, it's not just about recognizing objects but also interpreting their state-a person standing at the curb might step onto the street, a car signaling may change lanes, and a cyclist might swerve around a parked car. The robotaxi's software analyzes frame sequences and other sensor data to estimate object speed, direction, and likely next actions.

Maps, GPS, and Vehicle Localization

Standard satellite navigation isn't accurate enough for autonomy-GPS errors can be several meters, but a self-driving car needs much finer precision.

Most systems combine satellite navigation, inertial sensors, camera data, and highly detailed digital maps. The vehicle matches observed landmarks to map features to refine its position. High-precision maps include lanes, intersections, curbs, traffic lights, and other fixed road elements. These maps don't replace sensors, as real conditions constantly change, but they help the system understand the route's structure.

Predicting the Actions of Other Road Users

Once objects are identified, the robotaxi must predict what will happen next-a complex task since human behavior isn't always predictable.

The system models several possible movements for each object-like a car continuing, braking, or changing lanes, or a pedestrian waiting or stepping into the street. The motion planner evaluates these scenarios and selects a trajectory with a safety margin, slowing down for pedestrians at crosswalks even before they start moving.

The system continuously balances traffic rules, speed limits, obstacle distances, passenger comfort, and maneuver safety-aiming not only to avoid collisions but also to move predictably for others. This cycle of data collection, object recognition, prediction, and planning repeats constantly, making robotaxis advanced real-time computing systems.

How Waymo Robotaxi Works

Waymo is among the most renowned developers of self-driving taxis, using its proprietary Waymo Driver system for commercial driverless rides across several US cities. By August 2026, Waymo had completed over 20 million trips, with Waymo Driver accumulating more than 200 million miles of fully autonomous driving on public roads.

The Waymo Driver System

Waymo Driver combines hardware (cameras, LiDAR, radar, and onboard computers) and software. Its software fuses sensor data to solve four main tasks: localizing the vehicle, recognizing surrounding objects, predicting their next moves, and choosing the car's own maneuver.

For example, cameras recognize traffic lights and their signals, LiDAR measures distances to vehicles and objects, and radar estimates their speed. Combining these sources gives a fuller picture than any one sensor alone. Detailed digital maps provide additional context about road geometry, lanes, and intersections but don't replace real-time analysis-temporary obstacles must be detected by sensors during the ride.

Why Waymo Uses Cameras, Radar, and LiDAR

The main reason is redundancy. One sensor may be less effective in certain conditions, so others compensate for its limitations.

• Cameras: Excellent for colors and visual detail-recognizing signals, signs, and markings.
• LiDAR: Builds a precise 3D model and measures distances, less dependent on visible light-useful at night and in complex scenes.
• Radar: Accurately measures distance and speed, unaffected by rain, fog, or poor lighting.

This architecture increases cost and complexity but provides multiple independent ways to check surroundings-crucial for driverless passenger transport.

Where and How Waymo Robotaxis Operate

Waymo is gradually expanding its US commercial service. Fully autonomous rides are available in large cities such as Phoenix, San Francisco, Los Angeles, Miami, and others. In 2026, the company continued launching services in new markets and testing in additional regions.

Expansion is phased: before a full launch, cars collect data in a new city, maps are created and validated, and the system is tested under local road conditions. This allows the technology to achieve high reliability within a defined area before moving to the next city. In August 2026, Waymo announced preparations for launch in Munich, Germany.

Waymo vs. Traditional Driver Assistance Systems

Most ADAS help humans with individual tasks-lane keeping, following distance, or parking assistance. Even if the car can drive itself for long stretches, the driver remains responsible. By contrast, Waymo Driver is designed for autonomous control within its service area-passengers aren't expected to watch the wheel or pedals or be ready to intervene.

This changes the requirements: the system must handle not just routine driving, but also unusual situations-closed lanes, roadwork, poorly parked vehicles, or unpredictable road users. That's why Waymo focuses on sensor redundancy, detailed maps, large-scale testing, and gradual expansion, differing from Tesla's camera- and software-centric approach.

Tesla Robotaxi and Cybercab: A Different Approach

Tesla takes a distinct route, building robotaxis primarily around cameras, neural networks, and massive data sets from its fleet-rather than LiDAR, radar, and detailed maps.

As of August 2026, Tesla Robotaxi was already offering autonomous rides in Model Y vehicles in several Texas and Florida cities, including Austin, Dallas, Houston, Miami, Orlando, and Tampa, via a dedicated Robotaxi app. The specialized two-seater Cybercab is set to join the fleet in the future.

What Is Tesla Cybercab?

Cybercab is a two-seat electric vehicle designed from the ground up for autonomous passenger transport. Unlike adapting a regular car for robotaxi use, its design is optimized for driverless operation from the start-embodying Tesla's long-term vision of transportation built around autonomy.

Robotaxi and Cybercab are not the same thing: Robotaxi is Tesla's ride-hailing service, currently operating with Model Y; Cybercab is a purpose-built vehicle for this service and will be introduced gradually. This lets Tesla develop software capabilities before mass deployment of the new model. For passengers, the experience remains familiar: select a destination in the app, the system assigns a car, and after boarding, the vehicle completes the route autonomously within the available zone.

Why Tesla Focuses on Cameras and Software

Tesla's strategy centers on computer vision. Cameras around the car feed images to the onboard computer, where neural networks turn video streams into digital models of the environment, recognize markings, vehicles, people, and assess scene depth.

Tesla's algorithms analyze individual camera feeds, then merge them into a bird's-eye view with mapped infrastructure and 3D objects. The planning software then chooses the vehicle's trajectory. Unlike LiDAR-based systems, Tesla's neural networks derive most spatial information from visual data.

This approach scales well: cameras are compact and already installed on production Teslas. Data from the fleet is used to find complex scenarios and further train the computer vision system. Tesla emphasizes that its algorithms learn from real-world situations collected by its cars, and Robotaxi uses fleet learning to improve detection of markings, signs, and infrastructure.

The trade-off is much higher software demands: the neural network must correctly interpret 2D images even when objects are partly hidden, lighting is poor, markings are faded, or the road situation is unusual.

How Tesla Differs from Waymo

The contrast between Waymo and Tesla highlights two philosophies:

• Waymo: Multiple independent perception methods-cameras, LiDAR, radar-operating in mapped zones with detailed infrastructure data.
• Tesla: Relies on visual perception and AI-cameras provide input, neural networks interpret the scene, and software decides the next move.

Tesla's approach is potentially easier to scale: if autonomous software works reliably with a standard hardware suite, it's simpler to deploy a large fleet. Waymo's redundancy offers backup-if a camera struggles, other sensors can compensate.

Another key difference: Tesla's Full Self-Driving (Supervised) feature, available to owners, is not the same as a fully autonomous robotaxi. Tesla clearly states that FSD (Supervised) requires constant human attention and does not make a vehicle fully autonomous-the difference between supervised driving and commercial driverless rides is significant in terms of system responsibility.

Other Companies Developing Robotaxis

While Waymo and Tesla attract much attention, the robotaxi market is broader. Notable players include Zoox (an Amazon company) which has developed a fully symmetrical robotaxi not based on a traditional car design, expanding service in San Francisco and Las Vegas and testing in Austin and Miami as of 2026.

China is also a major market, with Baidu's Apollo Go deploying autonomous vehicles in multiple cities and expanding internationally-by August 2026, Baidu reported Apollo Go operating in 28 cities, with over 350 million autonomous kilometers logged (over 240 million fully driverless).

Apollo Go has its dedicated vehicle, the RT6, designed for true autonomy rather than as a modified production car. Another major developer is Pony.ai, scaling robotaxi operations in China and entering global markets. In 2026, its fleet neared 2,000 vehicles, with a partnership with Uber to expand in European cities.

There is no single technical standard yet. Some companies favor maximum sensor arrays and limited operating zones; others focus on computer vision and software scalability; some develop both an autonomous platform and a bespoke vehicle lacking a traditional driver's seat. Competition among these approaches will determine which model is most effective for mass urban transport: high sensor redundancy, universal computer vision, or a hybrid of several technologies.

Robotaxi Safety and When Will They Become Mainstream?

The main question about self-driving taxis is not whether they can drive themselves, but whether they can reliably handle rare and unpredictable situations. Mass adoption requires handling not just thousands of ordinary intersections, but also the one-off scenario the system may have never encountered.

Autonomous cars potentially have advantages over humans-they don't get tired, distracted, or speed due to haste, and can monitor multiple directions at once. But new risks arise: sensor limitations, algorithmic misinterpretation, mapping errors, software bugs, and unknown road scenarios.

Handling Complex or Unusual Road Situations

Routine trips on well-marked roads are predictable, but context-based interpretation is much harder-e.g., a traffic officer changing the flow, a construction worker blocking lanes, an unexpected maneuver by another driver, or a police signal the car has never seen. Even a minor accident can create a road configuration missing from the digital map.

In such cases, the autonomous system models various trajectories and usually opts for a cautious response-slowing down, stopping, or waiting for clarity. Some services also use remote support: an operator can provide extra information, though the car remains in control. Interacting with emergency responders remains a challenge-US regulator NHTSA in July 2026 flagged the need for correct response protocols in such scenarios.

Main Limitations of Today's Robotaxis

Modern robotaxis can't simply be sent unprepared to any country or address. Each system has a defined operational area-specific roads, cities, weather, and scenarios for which it's validated. Movement within the service area does not guarantee safe operation on snowy rural roads or cities with different traffic norms.

Challenges arise from heavy rain, snow, fog, blinding sun, or dirty sensors. Even if some sensors still work, the system must ensure there's enough data for safe operation. Rare road scenarios are also problematic-humans can sometimes deduce what's happening from context, but algorithms must base decisions on sensor data and accumulated experience.

This is why the quality of machine vision is a key element for autonomous transport. It's not enough to detect an object-the system must understand its location, type, movement, and potential behavior. For more on this, read our article on Machine Vision 2026: Key Trends, Technologies, and Applications.

How Safe Are Driverless Taxis?

Saying "robotaxis are safer than humans" is too general. Results depend on the system, region, driving conditions, and comparison group. However, as millions of autonomous kilometers accumulate, real-world safety data emerges. For example, by March 2026, Waymo vehicles had driven over 220 million miles autonomously. According to the company, the rate of severe or fatal crashes was 94% lower than comparable human drivers; crashes with any injury were 82% lower.

These figures show the technology's potential, but can't be generalized to all robotaxis-each system is different. Safety is not just about crash rates, but also about proper fail-safe stops, emergency responder interactions, real-time road changes, and safe actions if sensors fail. Regulators like the US NHTSA require manufacturers and operators to report certain crashes involving automated driving systems.

Barriers to Robotaxis in Every City

• Equipment and operating cost: Cars need computing systems, backup power, communication, and a full sensor suite-multiple LiDARs, radars, and cameras increase price and complexity.
• Software scalability: Millions of safe kilometers in one city don't guarantee reliability elsewhere-road markings, intersections, driver habits, signs, climate, and infrastructure differ city by city.
• Legislation: Fully driverless vehicle regulations vary by country and region, raising questions of legal responsibility, technical requirements, and safety certification.

For instance, in July 2026, Zoox received a temporary exemption from some US requirements, allowing commercial deployment of robotaxis with a unique design. Developing such vehicles requires updates to regulations historically built around a human driver.

Finally, driverless taxis must be economically viable-removing the driver cuts a major expense, but adds maintenance, remote support, sensor cleaning, charging, insurance, and fleet management costs.

When Will Self-Driving Taxis Become Commonplace?

Robotaxis are no longer just futuristic prototypes, but widespread adoption will likely happen city by city-not all at once. The best candidates are areas where systems can gather large data volumes and road conditions are well understood. Operators expand their service area only after proving reliability in a limited zone.

The transition to driverless transport will probably not be a sudden disappearance of drivers everywhere. Robotaxis will first become a normal ride option in certain major cities, then gradually expand as systems improve for tougher weather and road conditions.

The main metric will not be flashy demos or even a car completing a route without intervention, but rather the ability to repeat such trips millions of times with consistent safety and acceptable operating costs.

FAQ

1. Can a robotaxi drive completely without a driver?

   Yes. Level 4 robotaxis can operate without anyone behind the wheel, as long as they remain within their designated service area and road conditions meet system limits. This is how commercial driverless services like Waymo and others work. Level 5 vehicles, able to drive anywhere under any conditions, do not yet exist-most robotaxis are limited to specific cities or zones.

2. Which is safer-Waymo or Tesla?

   Direct comparison is difficult because the companies use different technologies, operate in different regions, and have accumulated different amounts of autonomous driving experience. Waymo uses cameras, LiDAR, radar, and high-precision maps for sensor redundancy; Tesla relies mainly on cameras and neural networks. Objective answers require comparable statistics across millions of trips in similar conditions-choosing a winner based solely on system design isn't correct.

3. Which companies are developing self-driving taxis?

   Beyond Waymo and Tesla, Zoox, Baidu (Apollo Go), Pony.ai, and others are actively developing robotaxi projects. They use different models-from modified production cars to purpose-built vehicles without a traditional driver's seat. The US and China lead in large-scale testing and commercial driverless zones.

4. When will self-driving taxis appear everywhere?

   There's no single date. The technology is already used commercially, but mass rollout depends on legislation, vehicle costs, safety, and systems' ability to adapt to new cities and weather. Adoption will likely be gradual-first becoming a familiar transport choice in major cities, then expanding as service areas grow and autonomy restrictions lessen.

Conclusion

Self-driving taxis are no longer a far-off technology. Robotaxis can independently recognize the road environment, predict the movement of others, plan routes, and transport passengers without a driver's constant involvement.

Waymo and Tesla represent two distinct paths: Waymo combines cameras, radar, and LiDAR with detailed maps and gradually expands tested service areas, while Tesla relies more on cameras, neural networks, and scalable software for its large vehicle fleet. Other companies, including Zoox, Baidu, and Pony.ai, are developing their own combinations of these approaches.

The main obstacle to mass adoption isn't a car's ability to drive a routine route, but ensuring consistent performance in rare and unusual scenarios. As autonomous systems accumulate real-world experience and prove their safety with statistics, driverless taxis will become a standard part of urban transport.