Inside Xpeng’s Robotaxi Plan: Level‑4 Autonomous Taxis, Turing AI Chips, and the Race With Tesla

The vehicle is built on Xpeng’s GX platform and engineered to meet Level‑4 autonomous‑driving standards. At this level, a car can operate without human intervention within specific conditions and geographic areas known as an operational design domain.

Unlike many autonomous‑vehicle programs that depend heavily on external suppliers, Xpeng developed the core technology itself—from computing hardware to the driving software. This approach is meant to give the company tighter control over cost, data, and system optimization.

Four Turing AI Chips Power the System

The robotaxi’s computing platform is built around four self‑developed Turing AI chips. Together they deliver roughly 3,000 tera operations per second (TOPS) of onboard computing power.

That processing capacity supports several key tasks required for autonomous driving:

• Perception: interpreting camera feeds and sensor data to understand the environment
• Prediction: anticipating the movements of vehicles, cyclists, and pedestrians
• Planning: determining the safest and most efficient driving path
• Vehicle control: translating decisions into steering, braking, and acceleration

Reports indicate the robotaxi uses a vision‑led approach to autonomy that avoids reliance on lidar sensors and high‑definition maps, instead depending primarily on camera data and AI models to interpret the road in real time.

The Role of Xpeng’s Vision‑Language‑Action Model

At the software level, the vehicles use Xpeng’s second‑generation VLA (vision‑language‑action) model. This system is designed to convert visual inputs from cameras directly into driving actions.

Traditional autonomous systems often separate perception, planning, and control into multiple stages. Xpeng’s VLA approach aims to streamline that pipeline by letting the AI model learn end‑to‑end behavior from large datasets of real driving video and scenarios.

The company frames this approach as part of a broader push into “physical AI,” where machine‑learning systems interact directly with the physical world through vehicles and robots.

Timeline: When Robotaxi Services Could Launch

Xpeng plans to start pilot robotaxi operations in the second half of 2026, initially in limited locations such as Guangzhou.

The company has already received permits to conduct public‑road testing of its Level‑4 robotaxi technology in the city.

However, large‑scale driverless deployment will likely take longer. Xpeng CEO He Xiaopeng has suggested that full autonomous driving could become achievable within one to three years, but that timeframe reflects company expectations rather than a guaranteed rollout.

In practice, commercial driverless services depend on several factors:

• Regulatory approval
• Safety validation through large‑scale testing
• Fleet operating costs
• Expansion of approved operating zones

Because of those constraints, a widespread driverless robotaxi service is more realistically viewed as a late‑decade milestone rather than an immediate launch.

Why This Matters in the Tesla Autonomy Race

Xpeng’s robotaxi program is also a strategic response to Tesla’s push into autonomous mobility.

Tesla’s Full Self‑Driving system relies heavily on camera‑based perception and large‑scale data collected from its vehicle fleet. Xpeng’s strategy mirrors parts of that approach while emphasizing deeper vertical integration.

Key similarities and differences include:

• Vision‑centric autonomy: Both companies emphasize camera‑based perception systems.
• AI‑driven software: Each uses large neural networks trained on driving data.
• Vertical integration: Xpeng is building its own AI chips and robotaxi vehicles, similar to Tesla’s hardware‑software strategy.

If Xpeng can deploy robotaxis at scale, it could shift competition in the EV market from simply selling cars to operating AI‑powered mobility platforms.

The Big Unknown: Scaling Driverless Mobility

Rolling a robotaxi off a production line is a milestone, but it does not guarantee widespread autonomous service. The hardest challenge remains proving that autonomous systems can operate safely and economically across complex urban environments.

Xpeng’s hardware and AI stack give it a credible foundation, but the outcome will depend on real‑world safety performance, regulatory approvals, and the economics of operating a robotaxi fleet. In other words, the technology race between Chinese EV makers and Tesla is increasingly about AI capability and deployment scale, not just electric vehicles themselves.