---
title: "Robotaxis in China — Baidu Apollo Go, Pony.ai and the World's Biggest Driverless Fleets"
date: 2026-09-17
category: AI in Action
site: NeuroAI
canonical: https://neuroai.site/a/na-robotaxi-china
language: en
---

# Robotaxis in China — Baidu Apollo Go, Pony.ai and the World's Biggest Driverless Fleets

> China runs the largest robotaxi (自动驾驶出租车) trials on Earth, led by Baidu's Apollo Go (萝卜快跑), Pony.ai and WeRide. This guide explains how they work, why Chinese cities moved faster than most, and what the safety and cost picture actually looks like.

In parts of Wuhan, you can hail a car with no driver and no steering wheel behind the wheel — just a screen and a seat. China's **robotaxi (自动驾驶出租车)** programs are no longer demos; they are paid, operating services moving real passengers. The leaders are Baidu's **Apollo Go (萝卜快跑)**, **Pony.ai**, and **WeRide**.

## Key takeaways

- China operates the **largest driverless ride-hailing fleets** in the world, concentrated in a handful of cities (Wuhan, Beijing, Guangzhou, Shenzhen).

- The three names: **Apollo Go (萝卜快跑)** from Baidu, **Pony.ai**, and **WeRide** — each with thousands of vehicles and growing paid zones.

- The model is **geofenced**: cars run safely inside mapped urban districts, not everywhere at once.

- China's edge is **policy + density**: friendly city permits and dense cities make the economics work faster than in sprawling suburbs.

- Safety data is improving but still **operator-reported** — treat headline numbers as directional, not independently audited.

## How a robotaxi works

Strip it down and a robotaxi is a self-driving stack on wheels:

- **Sensors** — cameras, radar, and lidar paint a 360° picture.

- **Perception** — software labels every object (pedestrian, bike, curb) many times a second.

- **Planning** — it decides speed, lane, and route under uncertainty.

- **Control** — it steers, brakes, accelerates.

The hard part is the *edge cases* — a jaywalker behind a bus, a puddle that looks like a hole. That is why deployment stays inside **geofences (运营区域)**: carefully mapped districts where the system has seen almost everything.

## The three leaders

- **Apollo Go (萝卜快跑)** — Baidu's service, the most aggressive on scale, with reports of tens of thousands of weekly orders in Wuhan at peak. It leans on Baidu's long-running Apollo autonomous platform.

- **Pony.ai** — a founder team with US roots, now China- and global-facing, running both robotaxi and freight pilots.

- **WeRide** — broad international ambition, operating pilots across multiple countries and chasing commercial robotaxi zones.

These sit alongside China's wider robotics push — from [humanoid factories](https://neuroai.site/a/na-jd-3-million-robots) to [Unitree's 18,000 humanoids](https://neuroai.site/a/na-unitree-18000-humanoids) — but robotaxis are the closest to everyday consumer use.

## Why China moved fast

Two forces converged:

- **City-level permission.** Chinese municipalities compete to host "smart city (智慧城市)" pilots and granted operating zones quickly.

- **Density economics.** Dense Chinese cities pack far more rides per square kilometer than US suburbs, so a fleet pays back faster.

The result is a feedback loop: more miles → more data → safer operation → wider zones.

## The cost and safety picture

The pitch is lower cost per ride than a human driver at scale, plus consistent, fatigue-free operation. The caveats:

- Most "safety" stats are **company-reported**, not independently audited.

- Scaling beyond geofences still depends on handling rare events reliably.

- Public trust is the real bottleneck — one bad incident slows a whole city's program.

## Honest limitations

This is an overview based on company disclosures and NeuroAI's reporting, not a safety audit. Fleet sizes, weekly-order counts, and city lists shift constantly and are often operator-reported; treat specific numbers as dated snapshots. "Largest in the world" refers to operating-scale driverless services as characterized in 2025–2026 industry coverage and may change. Lidar/sensor descriptions are generalized; exact hardware varies by generation and operator. No investment or safety conclusion should be drawn without current primary data.

## Sources

Baidu Apollo Go (萝卜快跑) operating disclosures; Pony.ai and WeRide public filings and pilot announcements; NeuroAI coverage of [JD's robot factories](https://neuroai.site/a/na-jd-3-million-robots) and [Unitree humanoids](https://neuroai.site/a/na-unitree-18000-humanoids); 2025–2026 industry reporting on Chinese robotaxi scale and city pilots.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-robotaxi-china
Free to quote with attribution and a link to the original.
