A robotaxi is a fully driverless, on-demand ride-hailing autonomous vehicle that operates at SAE Level 4 autonomy, navigating passengers or goods safely without a human driver present. Robotaxis rely entirely on AI software, sensors, and high-performance compute to perceive their environment and make real-time driving decisions.
The distinction isn’t a matter of degree. It is a categorical difference in who’s responsible for the vehicle.
In a Level 2 or Level 2++ vehicle, the human driver is legally and operationally responsible at all times while the AI assists. In a Level 3 vehicle, the AI handles driving in defined conditions but must alert the driver when it reaches the boundary of its capability. A human is always available as a fallback.
A robotaxi has no fallback. The vehicle operates in public, carrying passengers, with no human available to intervene. That single condition changes the technology requirements, the safety certification bar, and the operational model in every dimension.
The AI system must be able to handle not just common driving scenarios, but every situation the vehicle encounters within its ODD, including scenarios it wasn’t explicitly trained on. The safety architecture must be certified to automotive functional safety standards without relying on human oversight as a layer of protection. The fleet operations infrastructure must detect and respond to vehicle issues remotely. And the entire system must perform consistently across thousands of vehicles operating simultaneously in real-world conditions.
Robotaxis are being deployed across a growing range of industries and use cases, from urban ride-hailing services to the delivery of goods.
Ride-Hailing and Urban Mobility
Commercial ride-hailing networks are integrating robotaxis to offer scalable, on-demand transportation without the constraints of driver availability or shift limits. Multiple global mobility operators are currently scaling autonomous fleets in major cities around the world.
Autonomous Delivery
The same Level 4 vehicle platforms used for passenger transport are being adapted for last-mile delivery of groceries, packages, and prepared meals, extending the economic model beyond ride-hailing.
Healthcare and Accessibility
Robotaxis offer reliable, on-demand transportation for elderly individuals, people with disabilities, and others who cannot drive. Access to medical appointments, groceries, and daily errands becomes possible without depending on other people for rides.
Airport, Campus, and Geofenced Transit
Controlled environments such as airports, corporate campuses, and planned communities represent some of the earliest and most active robotaxi deployments. Defined routes, known infrastructure, and clear operational boundaries reduce complexity and accelerate regulatory approval.
Smart City Infrastructure
Municipalities are exploring how autonomous fleets can reduce total vehicle ownership, ease parking demand, and lower urban emissions, particularly when paired with all-electric powertrains and real-time traffic management systems.
Quick Links
Deploying robotaxis at scale requires solving regulatory, technical, and social challenges. Here are the primary barriers and the approaches the industry is using to address them.
Robotaxis demand a full-stack platform that spans in-vehicle compute, simulation, data center infrastructure, and AI models. NVIDIA provides every layer.
At the foundation is NVIDIA DRIVE Hyperion, a robotaxi-ready hardware platform built on the NVIDIA DRIVE AGX Thor system-on-chip. Thor delivers the AI compute performance needed for real-time perception, planning, and decision-making with the safety redundancy L4 autonomy requires.
In the data center, NVIDIA DGX and NVIDIA accelerated computing power the training pipelines that turn miles of sensor data into production-ready autonomous driving models. And with NVIDIA Omniverse with Cosmos, developers can generate the synthetic data and simulate the long-tail scenarios that are too rare or too dangerous to capture on public roads.
NVIDIA also provides open AI models and reference workflows that give AV developers the tools to build robotaxis.
The result is a full stack end-to-end platform, NVIDIA DRIVE, from training and simulation in the data center to inference and safety at the edge, that lets robotaxi developers build faster, validate more thoroughly, and deploy at scale with confidence.
The world's leading robotaxi companies run on NVIDIA. From the United States to China to Europe, virtually every major autonomous vehicle developer building toward commercial Level 4 deployment leverages NVIDIA technology.
In the United States, Waymo, Wayve, Avride, May Mobility, and Nuro build on NVIDIA technology, while Uber and Lyft are integrating NVIDIA-powered fleets into their global ride-hailing networks.
In China, Pony.ai, WeRide, and Momenta — among the most commercially advanced autonomous driving companies in the world — have chosen NVIDIA DRIVE Hyperion as the foundation for their next-generation platforms.
In Europe, Volkswagen, Mercedes-Benz, Autobrains, and Stellantis are advancing AV development and robotaxi-ready vehicle production on NVIDIA's platform.
Global automakers including BYD, Hyundai, Nissan, Geely, GM, and Toyota have also joined the ecosystem, making NVIDIA DRIVE Hyperion preferred reference architecture for robotaxi development. See list of partners at nvidia.com/autonomous-vehicles/partners.
Robotaxis rely on a suite of complementary sensors to build a complete, redundant picture of their environment, with the exact combination varying by company and approach.
No single sensor is sufficient on its own. Robotaxi stacks combine inputs from all of these sources through sensor fusion to build a reliable, real-time model of the world. ‘
NVIDIA DRIVE handles this at the compute level, processing data from cameras, LiDAR, and radar simultaneously through deep neural networks to enable safe L4 autonomy.
Cameras capture rich visual detail like lane markings, traffic lights, signs, and other road users. Most robotaxis use multiple cameras to achieve 360-degree coverage.
LiDAR (Light Detection and Ranging) fires laser pulses to generate precise 3D point clouds of the surrounding environment, excelling at measuring distance and detecting object shape and size, even in low light.
Radar uses radio waves to measure the speed and distance of objects, and is highly reliable in adverse weather conditions like rain, fog, and snow where cameras and LiDAR can struggle.
Ultrasonic sensors handle short-range detection for low-speed tasks like parking or detecting objects immediately adjacent to the vehicle..
Robotaxis are designed to meet a higher bar than human drivers. Human error accounts for over 90% of traffic accidents, and autonomous vehicles are built to eliminate the most common causes: distraction, fatigue, impaired judgment, and slow reaction times.
That said, safety in autonomous vehicles isn't a single feature, it's an engineering discipline. Rigorous safety requires redundant sensor arrays, fail-safe hardware architectures, billions of miles of simulation testing, and continuous validation against edge cases the vehicle may encounter in the real world.
NVIDIA's safety architecture is designed to meet ISO 26262 and SOTIF (Safety of the Intended Functionality) standards, the automotive industry's most rigorous functional safety requirements.
The robotaxi companies NVIDIA works with publish safety reports and maintain extensive internal validation programs. As the industry matures and fleets accumulate more miles, the data increasingly supports what the engineering was designed to achieve: autonomous vehicles that are meaningfully safer than human-driven ones.
A robotaxi is defined as a fully driverless, on-demand ride-hailing autonomous vehicle that operates at SAE Level 4 autonomy. This allows it to navigate passengers or goods safely without a human driver present. Learn more about in-vehicle computing.
NVIDIA DRIVE Hyperion™ is considered "robotaxi-ready" because it gives the world's automakers, AV developers, and mobility networks a common Level 4-ready foundation. It unites compute, sensors, safety software, and a global ecosystem to bring robotaxis from pilots to everyday transportation at scale. Learn more about in-vehicle computing.
The three-computer framework consists of NVIDIA DGX™ systems for training the AI-based stack in the data center, NVIDIA Omniverse™ running on NVIDIA OVX™ systems for simulation and synthetic data generation, and the NVIDIA AGX in-vehicle computer to process real-time sensor data for safety. Review the NVIDIA DRIVE Hyperion platform.
NVIDIA DGX systems are used for training end-to-end AI models in the data center by processing massive amounts of data. This helps autonomous vehicles learn to perceive, plan, and act more safely. Discover more about NVIDIA DGX for AI training.
NVIDIA Omniverse™ and Cosmos™ are used to build rich 3D digital twins that model real-world sensor data and physics, letting developers test and validate autonomous vehicles in a safe simulation environment. Visit the AV Simulation page for simulation details.
Alpamayo 2 Super is a 34 billion-parameter reasoning Vision-Language-Action (VLA) model specifically designed to support the sophisticated decision-making required for robotaxis and L4-ready autonomous vehicles.
NVIDIA Halos is a full-stack safety system designed to help robotaxi developers build, validate, and continuously improve autonomous vehicle safety. It spans hardware, software, and tools, so teams can address safety requirements across the entire development lifecycle rather than solving them in silos.
Deep dive into NVIDIA DRIVE AGX—the hardware and software platform to develop safe autonomous vehicles.
Safety is imperative to developing autonomous vehicles. Learn how NVIDIA is prioritizing vehicle, occupant, and pedestrian safety in our latest report.
NVIDIA automotive solutions offer the performance and scalability to design, visualize, develop, and simulate the future of driving.