EV44 — Baidu Apollo Go Fleet Failure, Wuhan (April 2026)

Evidence Note for CAM Governance White Paper

Compiled April 2026

---

Overview

On April 1, 2026, over 100 Baidu Apollo Go robotaxis simultaneously stalled on Wuhan’s elevated highway network during evening peak traffic, exposing a critical governance gap that existing AV regulations worldwide do not address: correlated fleet-level failures. Every existing regulatory framework — in China, the US, and the EU — was designed around individual vehicle defects and single-vehicle incidents. The Wuhan event demonstrated that when CAM operates at scale, a single infrastructure or software failure can cause dozens or hundreds of simultaneous incidents across an entire urban network.

---

Incident Details

• Date: April 1, 2026
• Location: Wuhan, China — Second and Third Ring Road elevated expressways
• Fleet affected: Over 100 Baidu Apollo Go robotaxis simultaneously
• Duration: Vehicles stranded for up to 90 minutes
• Cause (official): “System malfunction” — per Wuhan traffic police; Baidu provided no detailed public explanation
• Technical assessment (industry analysts): Probable cloud dispatch infrastructure failure rather than individual vehicle defects; likely triggered fleet-wide safety self-check mechanism

Sequence of Events

1. Mass simultaneous stall of 100+ vehicles across elevated expressways during peak hour
2. Vehicles failed to execute expected fail-safe manoeuvres (should have guided to safe stopping locations)
3. Multiple rear-end collisions caused by following drivers encountering stationary AVs without warning
4. Passengers trapped inside vehicles during stall period — unable to exit safely on elevated highway
5. Traffic gridlock across elevated highway network
6. No fatalities reported; all passengers eventually exited safely

---

Precursor: Baidu Apollo RT6 Incident, Zhuzhou (December 2025)

• Date: December 2025
• Location: Yanjiang Road, Zhuzhou, Hunan Province
• Time: Approximately 9:00 AM
• Vehicle: Baidu Apollo RT6 (6th-generation AV, jointly developed with JMC), operated by Hello (formerly Hellobike)
• Casualties: Two pedestrians struck; both hospitalised in intensive care
• Status: No detailed root cause analysis published

This precursor demonstrates that the Wuhan fleet event did not occur in isolation — Apollo Go operations had already produced a serious pedestrian injury incident three months earlier.

Source: carnewschina.com https://carnewschina.com/2025/12/06/baidus-robotaxi-causes-accident-in-a-central-city-of-china-leaves-two-in-intensive-care/

---

Scale Context: Baidu Apollo Go Operations in Wuhan

• Wuhan is Baidu’s largest robotaxi market; Apollo Go has operated there since 2021
• By mid-2025, Baidu reported 2.2M fully driverless rides in Q2 2025 alone (+148% YoY)
• Cumulative total: 14M+ rides in China by August 2025
• Fleet scale in Wuhan makes city-level infrastructure failure scenarios operationally plausible

---

The Governance Gap: Fleet-Level vs. Individual Vehicle Regulation

What existing regulations address:

• Individual vehicle defects (hardware failure, software bug in single unit)
• Individual vehicle incidents (collision, emergency stop)
• Individual vehicle recalls (OTA updates affecting fleet, triggered by single-vehicle findings)

What existing regulations do NOT address:

• Correlated failure: When a single cloud, dispatch, or software event affects an entire operating fleet simultaneously
• Cascading safety mechanisms: When fleet-wide safety protocols designed to protect individual vehicles create system-level gridlock
• Urban network effects: When dozens of simultaneous stationary AVs create secondary incidents (rear-end collisions by human drivers)
• Passenger egress on infrastructure: No protocol existed for extracting passengers from stationary AVs on elevated highways without shoulders
• Communication failure: No public alert system to warn following traffic of mass simultaneous stalls ahead

Why this matters for governance

The Wuhan incident is architecturally distinct from previous AV incidents (Cruise pedestrian drag, Waymo intersection standoffs, Xiaomi NOA crash). Those were individual vehicle events triggering individual regulatory responses. The Wuhan event is a system-level failure — the AV fleet, the dispatch infrastructure, and the urban highway network behaved as a single interconnected system, and that system failed simultaneously for over 100 units.

This represents a new failure mode category: Fleet Coherence Failure — distinct from the five failure modes in the current CAM paper framework and potentially warranting treatment as a sixth, or as a sub-mode of Allocation failure.

---

Governance Implications

For city-level CAM governance:

1. Fleet coherence standards: Before authorising large-scale robotaxi operations, cities must require operators to demonstrate fleet-level failsafe architecture — specifically: what happens if cloud dispatch is unavailable? What happens if a safety-check firmware event affects all vehicles simultaneously?

2. Graduated deployment: Fleet size caps during operational learning phase prevent single-event failure from becoming city-wide gridlock. Wuhan’s large fleet enabled the scale of disruption.

3. Urban network integration: AV dispatch systems must integrate with city traffic management centres in real time. When 100+ vehicles stall simultaneously, traffic signal systems, variable message signs, and emergency responder dispatch must be automatically triggered.

4. Passenger egress protocols: Operations on elevated highways, tunnels, and other infrastructure without safe pedestrian egress require bespoke passenger safety protocols — including communication, emergency services integration, and physical egress plans.

5. Regulatory framework update: China’s post-Xiaomi April 2025 guidelines addressed individual vehicle OTA governance. The Wuhan incident reveals these rules are insufficient for fleet-scale operations. A fleet-level equivalent is needed.

---

Comparison: US Context

The closest US analogue is the series of Waymo standoff incidents (2023–2024) where individual Waymo vehicles became confused and stopped in intersections. These were individual vehicle events affecting traffic locally. The Wuhan event is an order of magnitude larger and architecturally different — it is a cloud infrastructure event, not a vehicle AI event.

No US regulatory framework addresses fleet-level correlated failure scenarios. NHTSA FMVSS standards are vehicle-by-vehicle. Senator Markey’s March 2026 RAO investigation focused on remote operator qualifications, not fleet infrastructure architecture. This is an unaddressed gap at all regulatory levels.

---

Key Metrics

Metric	Value
Vehicles simultaneously affected	100+
Duration of stall	Up to 90 minutes
Secondary incidents	Multiple rear-end collisions
Fatalities	0
Injuries	Not reported
Regulatory response	None published as of April 2026

---

Sources

• TheNextWeb — “Over 100 Baidu robotaxis froze mid-traffic in Wuhan” https://thenextweb.com/news/baidu-apollo-go-robotaxi-wuhan-mass-malfunction
• CNN — “Baidu robotaxis stranded in Wuhan after system failure” https://www.cnn.com/2026/04/02/china/baidu-robotaxi-wuhan-system-failure-intl-hnk
• TechCrunch — “System failure paralyzes Baidu robotaxis in China” https://techcrunch.com/2026/04/01/system-failure-paralyzes-baidu-robotaxis-in-china/
• CNBC — “Baidu robotaxis caused crashes in Wuhan” https://www.cnbc.com/2026/04/01/baidu-robotaxis-caused-crashes-in-wuhan-reports.html
• SCMP — “Baidu’s robotaxi breakdown in Wuhan strands riders, raises safety concerns” https://www.scmp.com/tech/article/3348654/baidus-robotaxi-breakdown-wuhan-strands-riders-raises-safety-concerns

---

Document Status: Compiled April 2026. Incident occurred April 1, 2026; regulatory response (if any) expected in subsequent weeks. Check for follow-up.