The Chinese model built in Guangzhou aims for L4 autonomous driving, four Turing AI chips and a platform without LiDAR or HD maps.

The distance between an autonomous driving prototype and a truly industrializable service isn't measured solely in kilometers traveled on the road. It's measured in the ability to transform software, sensors, on-board computing, vehicle platform, operations, and maintenance into a repeatable product. It's on this ground that XPENG extension has positioned the rollout of its first Robotaxi off the production line in Guangzhou as a step from technology demonstration to mass production.

According to the company, this is the first time in China in which a car manufacturer achieves series production of a Robotaxi through a development full stack This statement should be read in the competitive context of the Chinese smart electric vehicle market, where manufacturers, digital platforms, and mobility operators are trying to understand which combination of hardware, software, and service model can make autonomous driving sustainable beyond the experimental phase.

The new vehicle is based on the platform XPENG GX and is designed to meet standards of Level L4 autonomous drivingIn practical terms, level L4 indicates a system capable of operating without human intervention within defined operating conditions, such as geographic areas, road scenarios, and established functional limits. The challenge, therefore, is not simply to get the driverless vehicle moving, but to do so with continuity, safety, predictable costs, and industrial governance suitable for a public or semi-public service.

XPENG's choice is also significant because it concentrates a large portion of the core competencies within the vehicle: chips, software stack, operating system, electrical and electronic architecture, onboard platform, and user interface. This strategy reduces dependence on external suppliers, but increases industrial complexity. For a robotaxi, product maturity isn't just about autonomy: it also includes diagnostics, updates, fleet management, passenger experience, and integration with urban digital ecosystems.

From the road test to the production line in Guangzhou

The company's path shows a rather clear sequence. In January, XPENG's Robotaxi obtained permits for road tests in Guangzhou dedicated to connected intelligent vehicles, entering the phase of routine public testing for L4 applications. In March, a dedicated Robotaxi business unit was established, tasked with coordinating product definition, development activities, and testing. research and development, testing and operation.

Moving to the production line does not automatically equate to full commercialization. XPENG plans to launch pilot operations in the second half of 2026, with the aim of validating three different dimensions: technical feasibility, user acceptance, and sustainability of the entire system. business modelFull operation without a safety officer on board is set as a target for early 2027, a deadline that will also depend on authorizations, service maturity, and local operating conditions.

According to Reuters, President Brian Gu has estimated that XPENG could produce several hundred to several thousand robotaxis over the next twelve to eighteen months. This scale is still far from mass deployment, but it's sufficient to shift the focus from demonstration to operational management. For an industry that has for years alternated between high expectations and cautious revisions, the number of vehicles in service will be less important than their actual availability, cost per kilometer, and ability to operate in complex urban environments.

The industrial point is precisely this: a robotaxi is not a traditional car with advanced software added later. It is a product-service that requires integrated design from the outset. The vehicle must be robust, upgradeable, easily monitored, and designed for intensive use. Even the passenger compartment, seemingly marginal to autonomy, becomes part of the service economy, as it impacts users' comfort, trust, and willingness to ride in a driverless vehicle.

Pure Vision and VLA 2.0: The Challenge Without LiDAR

The most relevant technological choice concerns the absence of LiDAR and high definition maps. XPENG declares to adopt a solution pure vision, in which the decision-making process is driven by the large end-to-end model VLA 2.0. Instead of relying on a three-stage Vision-Language-Action chain with intermediate language translation steps, the architecture aims to compress the system response time below 80 milliseconds.

Pure vision is an engineering and industrial challenge. On the one hand, it can reduce costs, calibration complexity, and dependence on expensive sensors; on the other, it requires very robust models, extensive training data, and a high capacity to generalize to unexpected situations. In an urban context, the challenge isn't recognizing an ordered scenario, but managing constant exceptions: uncertain pedestrians, cyclists, construction sites, mixed traffic, irregular signage, weather conditions, and local behavior.

The vehicle is powered by four AI Turing chips owners, for effective computing power on board 3.000 TOPSThis data is significant because it indicates an architecture designed to process a substantial portion of decisions locally. In robotaxis, latency is not an abstract parameter: it determines the time between perception, interpretation, and action. Reducing it increases operational margin in cases where the vehicle must react quickly, without waiting for remote processing or depending on perfect connectivity.

This doesn't eliminate the role of the cloud and management platforms, but it does redefine the system's focus. The fleet will still need to send data, receive updates, and be monitored and coordinated. However, on-board autonomy remains key to functional safety and geographic scalability. XPENG claims that VLA 2.0 can support urban generalization capabilities, making it easier to deploy in different cities and even in cross-border scenarios. This is an ambitious goal, requiring extensive testing and specific authorizations.

“The true value of intelligent driving lies not just in capability, but in offering greater efficiency and a more relaxed experience.”

The sentence of He Xiaopeng, Chairman and CEO of XPENG, summarizes the shift in perspective that many manufacturers are trying to bring to autonomous driving. It's no longer enough to present the technology as an exercise in computational power. It must be demonstrated that it produces a measurable benefit for users, cities, and operators: more predictable times, perceived safety, better fleet utilization, and costs compatible with a repeatable service.

A cockpit designed to validate the service

XPENG's Robotaxi is not only described through sensors and models of Artificial intelligenceThe company also focuses on the interior, featuring privacy glass, gravity comfort seats, rear entertainment screens, and an integrated voice assistant. These elements may seem like accessories, but in a driverless service, they become part of building trust. The user must feel controlled, protected, easy to use, and clear interaction.

The presence of voice commands and multimedia functions indicates that XPENG envisions the Robotaxi as a passenger-controlled travel environment, not just a simple automated shuttle. This approach is consistent with a potentially premium service level, or at least a superior positioning compared to traditional public transport. It remains to be seen whether this configuration will be sustainable on a large scale, where maintenance, cleaning, damage, and intensive use cycles impact operating costs.

Another important step concerns the ecosystem. XPENG plans to open its own Robotaxi SDK, while Amap will become the ecosystem's first global partner. This detail is important because no robotaxi exists solely within the manufacturer's perimeter. It requires maps, reservations, routing, payments, customer support, anomaly management, and interfaces with local authorities. The controlled openness of the SDK can be used to build applications and integrations, but it also raises issues of liability, security, and quality of service.

Commercialization will therefore depend on a balance between proprietary integration and external collaboration. A system that is too closed can slow adoption; one that is too open can increase control complexity. For a robotaxi operator, the value lies not only in the single vehicle, but in the ability to orchestrate many, updating them seamlessly and ensuring consistent standards. This is where autonomous mobility intersects with approaches closer to industrial software than traditional automotive sales.

Physical AI Links Cars, Humanoid Robots, and Flight

The Robotaxi is presented by XPENG as one of the flagship products of its ecosystem Physical AI, the same technological horizon that includes the humanoid robot IRON and the flying car. The term refers to the application of AI to the physical world, where models don't just generate text, images, or predictions, but must also perceive, decide, and act in real environments. This is a more difficult step, because any error can have immediate material consequences.

Sharing technological foundations between vehicles, robots, and air mobility systems can generate economies of learning. Chips, models, compilers, software stacks, and simulation tools can be reused, at least in part, across different platforms. However, each domain has its own unique requirements. A robotaxi must navigate urban traffic and passengers; a humanoid must navigate spaces built for people; an aircraft requires certification, redundancy, and aeronautical regulations. Talking about a common platform doesn't mean erasing these differences.

For the automotive industry, the XPENG case illustrates a broader trend: the intelligent electric vehicle is becoming a mobile computing platform. The competition isn't just about range, battery life, or design, but the ability to transform the vehicle into a hub for data, services, and continuous updates. In this scenario, the mass production of a Robotaxi could become a testbed for architectures that could power other forms of automated mobility in the future.

The economic question remains. Robotaxis promise to reduce driver costs, but require significant investments in hardware, software, insurance, remote supervision, cleaning, maintenance, and compliance. Success will depend on demand density, local regulation, and the ability to operate at high utilization rates. In cities with heavy traffic, clear regulations, and digitally mature users, the model may find favorable conditions; elsewhere, the transition could be much slower.

The Guangzhou rollout doesn't end the autonomous driving game, but it does change the level of verification. From now on, the challenge will not simply be to demonstrate that a vehicle can drive itself, but to prove that a fleet can do so in a reliable, accepted, and sustainable way. For XPENG, the goal is to shorten the cycle between development and commercial operation; for the industry, the test will be to understand whether mass production can truly transform autonomy from a technological promise to a mobility infrastructure.

La China It remains one of the most closely watched laboratories for this transition, thanks to the combination of the automotive market, digital platforms, urban policies, and industrial competition. But the real test will be replicating the experience beyond the first authorized perimeter, maintaining security and service quality. It is in this ability to move from a single announcement to operational continuity that the value of the Artificial intelligence applied to physical mobility.

Here are three insights that might interest you:

Thus the Swiss government intends to allow automated driving
Robotaxi: Nissan, Uber, and Wayve Experiment with Urban AI
Sweden, the revolution of electrified roads for moving vehicles