# Apollo Go Wuhan

The largest single-city robotaxi deployment in the world and the deployment where what failure-at-scale looks like first became publicly documented. Baidu's Apollo Go operates more than 1,000 fully-driverless vehicles in Wuhan, China, with 24/7 operation since 2024 and the largest cumulative operating volume of any robotaxi service globally. On March 31, 2026, more than 100 of those vehicles simultaneously stopped responding to commands during evening rush hour, freezing on city streets and elevated highways in what authorities described as a system malfunction.

## The short version

Apollo Go Wuhan is the verified-scale case study in robotaxi deployment. The operation has the largest single-city fleet, the highest cumulative ride volume, and the cleanest operating record for sustained fully-driverless commercial service of any robotaxi globally. It is also the operation where the first publicly-documented fleet-wide system failure occurred, providing the first real-world data on how robotaxi operations behave when central infrastructure fails. Both stories are part of what robotaxis-at-scale actually look like.

## The operating site

Wuhan is the capital of Hubei province in central China, a city of approximately 11 million people. Apollo Go received the necessary local-government and central-government permits to operate fully driverless robotaxi service in Wuhan beginning 2024. The Wuhan operation is the company's largest commercial deployment by fleet size and accounts for the majority of Apollo Go's fully-driverless operating volume. Apollo Go also operates in Beijing suburbs and additional Chinese cities; Wuhan is the reference site for full-scale operation.

The vehicles operating in Wuhan include the fifth-generation Apollo platform and, increasingly, the purpose-built sixth-generation RT6 robotaxi. The RT6 is reported to cost approximately $27,500 per unit to manufacture (approximately 200,000 RMB), roughly half the cost of the prior generation. Apollo Go CEO Robin Li stated at Baidu World 2022 that the cost reduction would enable deployment of tens of thousands of vehicles across Chinese mega-cities, with operations claimed profitable on a unit-economics basis starting 2025.

## Scale milestones

The Apollo Go program produces unusually well-disclosed scale metrics through Baidu's earnings releases. Verified data points across the program:

2013: Baidu enters autonomous driving research and development.
2022: Baidu announces sixth-generation RT6 robotaxi at Baidu World; targets 1,000 RT6 deployment in Wuhan within the year.
2024: 24/7 fully-driverless operation achieved in Wuhan.
Q4 2025: 3.4 million fully-driverless rides delivered (200%+ year-over-year growth).
October 31, 2025: Cumulative autonomous driving mileage exceeds 240 million kilometers; cumulative rides exceed 17 million globally.
February 2026: Apollo Go services deployed in 26 cities globally; cumulative orders exceed 20 million per Baidu earnings release.
Wuhan single-city fleet: more than 1,000 vehicles operating fully driverless.

For context: Apollo Go's cumulative ride volume and single-city fleet size place it as the largest robotaxi operator globally by these metrics, ahead of Waymo (3,000+ fleet across 10 US cities, ~500,000 weekly rides as of March 2026).

## International expansion

Apollo Go received its first international autonomous driving test license in Hong Kong in November 2024, the first foreign autonomous-vehicle program permitted in Hong Kong. International expansion has accelerated through 2025-2026:

March 28, 2025: Apollo Go and Dubai Roads and Transport Authority (RTA) sign strategic partnership for robotaxi deployment in Dubai. Initial fleet target: 100 vehicles by end of 2025, scaling to no less than 1,000 by 2028. This is Apollo Go's first international fleet deployment outside mainland China and Hong Kong.

December 2025: Uber and Lyft both announce partnerships to bring Apollo Go vehicles to UK roads, with London as the initial deployment target for H1 2026. Uber's pilot expected to use RT6 vehicles; Lyft acquisition-related testing planned with 50-vehicle initial fleet.

March 2026: Briefly suspended UAE operations early March 2026 for regulatory compliance, resumed March 10. Officially launched fully-driverless commercial operations in Dubai March 30, 2026.

The international expansion makes Apollo Go the first Chinese robotaxi operator to achieve commercial driverless service outside China at meaningful scale.

## The Wuhan mass system failure, March 31, 2026

The largest publicly-documented operational failure in robotaxi history occurred on Apollo Go's Wuhan fleet on March 31, 2026 at approximately 8:57 PM local time. Verified incident details:

More than 100 Apollo Go robotaxis simultaneously stopped responding to commands across multiple districts of Wuhan. Vehicles came to a complete halt in their current lanes, including in middle lanes of ring roads with traffic streaming past on both sides, on elevated expressways, and at intersections. Vehicles did not pull over to the side of the road; they did not activate an emergency protocol; they simply stopped where they were. Passengers were trapped inside immobilized vehicles and called police; hundreds of distress calls were received.

Wuhan traffic police issued a statement on Weibo describing the cause as a "system malfunction" with further investigation underway. Authorities confirmed no injuries were reported and all passengers exited their vehicles safely. One video reportedly showed the outage causing a secondary highway collision, though Wuhan police confirmed no injuries.

Operations resumed the same evening after Baidu identified and resolved the cause. The incident occurred during a period when Apollo Go was actively expanding internationally; the timing drew immediate scrutiny from UK regulators ahead of the planned London pilot.

This is the first publicly-documented case of a fleet-wide central-infrastructure failure in a commercial robotaxi service. The category significance is that, while individual robotaxi incidents (vehicles striking pedestrians, vehicles failing to detect specific hazards) have been documented in multiple operations, a centrally-managed-fleet single-point-of-failure had not previously been demonstrated at scale.

## Earlier incident, August 2025 Chongqing

A separate Apollo Go incident in Chongqing in August 2025 also entered the verified record. An Apollo Go robotaxi carrying a single woman passenger drove into a construction pit, bypassing barriers and warning signs before dropping into the trench. The passenger was rescued unharmed by nearby residents using a ladder. The incident illustrated a known autonomous-vehicle vulnerability: difficulty detecting large, irregular road hazards that fall outside the system's training datasets.

The Chongqing and Wuhan incidents together establish a pattern that autonomous-vehicle researchers have called "fails in completely new ways." Autonomous systems fail in modes that human-driven systems would not, and regulatory frameworks designed for human-driver-error patterns are not structured to address these failure modes.

## What this deployment proves

Three things, each of which were live questions for robotaxi-at-scale before Apollo Go Wuhan reached current scale:

That a robotaxi fleet can sustain 24/7 fully-driverless commercial operation at thousand-vehicle scale in a major city for multiple years. Apollo Go Wuhan answered the operational question that Western robotaxi programs were still proving through 2024-2025.

That a robotaxi operator can scale internationally on the strength of domestic operating record. Apollo Go's Dubai, Hong Kong, and pending London deployments are validated by the Wuhan and broader Chinese operating record. The international expansion pattern is being followed by Western operators (Waymo to London) but Apollo Go is the first to execute it.

That fleet-wide infrastructure failures are a real risk class for centrally-managed autonomous fleets. The Wuhan incident is, paradoxically, a demonstration of the verified-record discipline: the failure was acknowledged publicly, documented in real time, investigated, and resolved within hours. The verified-record discipline did not prevent the failure; it shaped how the failure was processed by regulators, passengers, and the broader public.

## What to watch

Three things matter most for what Apollo Go Wuhan becomes in the next eighteen months.

First, whether the March 31 incident causes any of the planned international launches (London via Uber and Lyft; expanded Dubai) to slip or be rescoped. Western regulators have an existence-proof of fleet-wide failure to consider that they did not have before March 31, 2026.

Second, whether the technical root cause of the Wuhan incident becomes public. As of late May 2026, Baidu has not published a detailed post-incident report. A robotaxi-at-scale operator publishing detailed root-cause documentation for a fleet-wide failure would set the industry norm; whether Baidu does this is itself a verified-record-discipline question.

Third, whether other robotaxi operators (Waymo, Pony AI, WeRide, Tesla Robotaxi) experience similar central-infrastructure failures through 2026. The Wuhan incident may represent a unique-to-Apollo-Go architecture issue or it may represent a category-wide risk pattern. The next twelve months of operating data across operators will indicate which.

*The verified record is below. Numbers reflect Baidu earnings releases, Wuhan traffic police statements, and primary-source coverage as of late May 2026. The March 31, 2026 Wuhan incident is treated as a verified incident in DEPLOY's incident registry.*

## Operational data

| Field | Value |
|---|---|
| Operator | Baidu (Apollo Go is a Baidu division) |
| Service area | Wuhan, Hubei Province, China (~11M population) |
| Wuhan fleet | 1,000+ vehicles, fully driverless |
| Service type | 24/7 fully-driverless commercial (since 2024) |
| Vehicle | Apollo RT6 (6th-generation purpose-built robotaxi) |
| Vehicle cost (claimed) | ~$27,500 / ~200,000 RMB per unit |
| Q4 2025 rides | 3.4 million fully-driverless (200%+ YoY) |
| Cumulative rides (Feb 2026) | 20 million+ across 26 cities globally |
| Cumulative autonomous km (Oct 2025) | 240 million+ |
| Permit (China) | First fully-driverless commercial permit in Wuhan + Chongqing (Aug 2022) |
| International (Dubai) | Commercial launch March 30, 2026 (target 100 by end-2025, 1,000+ by 2028) |
| International (Hong Kong) | First foreign AV permit (Nov 2024) |
| International (London, planned) | Uber + Lyft pilots, H1 2026 / 50-vehicle Lyft initial |
| Unit economics (claimed) | Per-vehicle profitability reached 2025 in Wuhan geofence |
| Major incident (Wuhan) | March 31, 2026, 8:57 PM local. ~100+ vehicles simultaneously frozen on streets and highways. No injuries. Operations resumed same evening. |
| Earlier incident (Chongqing) | August 2025. Vehicle drove into construction pit with passenger; passenger rescued unharmed. |

_Registry detail: [/deployments/apollo-rt6-wuhan](/deployments/apollo-rt6-wuhan.md) is the canonical machine-readable record._

## Related entities

- [Baidu on the registry](/companies/baidu.md)

## Related comparisons

- [Waymo vs Tesla](/waymo-vs-tesla.md)
- [Pony AI vs WeRide](/pony-ai-vs-weride.md)
- [Waymo vs Cruise](/waymo-vs-cruise.md)

## See also

- [All marquee deployments](/deployments/marquee.md)
- [Methodology](/methodology.md). What we count, how we verify.
- [Glossary](/glossary.md). Strict definitions used across the registry.

_Canonical URL: /deployments/marquee/apollo-go-wuhan_