California Autonomous Vehicle Testing
California is the only U.S. state that requires autonomous vehicle companies to publicly disclose every “disengagement” — moments when the self-driving system hands control back to a human. This data covers the December 2024 - November 2025 report period.
Source: CA DMV Disengagement Reports · Last updated 2026-04-15
Total Miles Driven
8.3M
On California public roads
Companies Testing
12
9 currently active
Best MPI
63,415
Waymo
Total Disengagements
1,850
Across all companies
Key Insights
- →Waymo dominates California AV testing with 5.2M miles driven and an industry-leading 63,415 miles per disengagement — over 30x better than most competitors.
- →Cruise, once the #2 AV operator in California, suspended all operations after an October 2023 pedestrian-dragging incident. Their driverless testing permit was revoked by the DMV.
- →Delivery-focused companies like Nuro (2,000 MPI) and Gatik (1,493 MPI) show strong reliability despite operating in more structured, lower-speed environments.
- →California remains the gold standard for AV transparency — no other state requires this level of public disclosure, making this dataset uniquely valuable for safety analysis.
Company Rankings by Miles per Disengagement
Higher miles per disengagement (MPI) indicates better autonomous performance. Companies are ranked from most to least reliable.
| # | Company | Miles Driven | Disengagements | MPI | Vehicles | Permits | Status |
|---|---|---|---|---|---|---|---|
| 1 | Waymo | 5,200,000 | 82 | 63,415 | 700 | Testing (with driver)Testing (driverless)Commercial Deployment | improving |
| 2 | Nuro | 300,000 | 150 | 2,000 | 40 | Testing (with driver)Testing (driverless) | improving |
| 3 | Zoox | 1,200,000 | 800 | 1,500 | 110 | Testing (with driver)Testing (driverless) | improving |
| 4 | Gatik | 100,000 | 67 | 1,493 | 15 | Testing (with driver) | improving |
| 5 | Pony.ai | 400,000 | 333 | 1,201 | 50 | Testing (with driver) | stable |
| 6 | WeRide | 150,000 | 167 | 898 | 25 | Testing (with driver) | stable |
| 7 | Aurora | 200,000 | 250 | 800 | 30 | Testing (with driver) | improving |
| 8 | AutoX | 100,000 | 167 | 599 | 20 | Testing (with driver) | stable |
| 9 | Mercedes-Benz | 50,000 | 100 | 500 | 10 | Testing (with driver) | stable |
| 10 | Motional | 80,000 | 200 | 400 | — | Testing (with driver) | paused |
| 11 | Cruise | 500,000 | 2,500 | 200 | — | Testing (with driver) | suspended |
| 12 | Apple (Project Titan) | 10,000 | 134 | 75 | — | Testing (with driver) | cancelled |
What Is a Disengagement?
A disengagement occurs when the autonomous driving system either deactivates itself or the human safety driver takes manual control of the vehicle. Common reasons include:
- System detected a situation outside its operational design domain
- Sensor malfunction or degraded perception
- Unexpected road conditions (construction, debris, etc.)
- Safety driver perceived an imminent risk
- Software fault or planning failure
The key metric is miles per disengagement (MPI)— a higher number means the system can drive farther without needing human intervention. Waymo's 63,415 MPI means their vehicles average over 63,000 miles between disengagements.
California AV Permit Types
🧑✈️ Testing with Driver
A trained safety driver must be present in the vehicle at all times. This is the most common permit type and the entry point for all AV companies testing in California.
🤖 Driverless Testing
Vehicles can operate without a human driver, but a remote operator must be able to take control. Requires additional safety certifications and insurance.
🚀 Commercial Deployment
Companies can offer autonomous rides or deliveries to the public for a fee. Currently only Waymo holds this permit in California.
Why California's Data Matters
California is the only state in the U.S. that requires autonomous vehicle companies to publicly report disengagement data. While states like Arizona, Texas, and Florida have AV-friendly policies, none mandate this level of transparency.
This makes California's dataset uniquely valuable — it's the closest thing we have to a standardized safety benchmark for self-driving technology. However, it has limitations: companies self-report, definitions of disengagement can vary, and miles driven in controlled vs. challenging environments aren't distinguished.