What Is the Tesla Robotaxi Crash Rate?
As of December 2025, Tesla's robotaxi fleet in Austin, Texas is achieving approximately 92,500 miles per incident (MPI). This means that, on average, the fleet drives about 92,500 miles between each incident reported to NHTSA under Standing General Order 2021-01.
This figure represents the most recent completed interval between two consecutive NHTSA-reported incidents. It is derived from the fleet size (tracked daily via robotaxitracker.com), an estimated 115 miles per vehicle per day (disclosed by Tesla in Q3 2025), and the dates of NHTSA-reported crashes.
It is important to note that this number changes with every new incident. For the most current figure, view the live data on our main tracker.
How Is the Crash Rate Improving?
The most significant story is not any single MPI number, but the trend. Tesla's robotaxi safety has been improving at an exponential rate, with MPI approximately doubling every 79 days.
When Tesla launched its unsupervised robotaxi service in Austin in June 2025, the early intervals between incidents were short, often just a few thousand miles. By late 2025, intervals had grown to tens of thousands of miles, and the trajectory continues upward.
This exponential improvement pattern is consistent with a system that learns from real-world data. Each incident provides training signal. Each software update improves the driving model. The result is a compounding safety curve that, if sustained, projects Tesla's fleet crossing human-driver safety benchmarks in 2026.
The exponential trend fit has an R-squared value close to 1.0, indicating that the model explains the observed data well. However, past performance does not guarantee future improvement, and the trend should be monitored continuously.
How Does Tesla Compare to Human Drivers?
Comparing robotaxi crash rates to human driver crash rates is more complicated than it appears. There are several commonly cited benchmarks for human drivers:
| Benchmark |
Miles Per Incident |
What It Measures |
| Police-reported crashes |
~500,000 |
Crashes serious enough for a police report |
| Insurance claims |
~300,000 |
Crashes where an insurance claim was filed |
| All crashes (estimated) |
~100,000-200,000 |
Including unreported minor incidents |
At 92,500 MPI, Tesla's robotaxi fleet is currently below all of these human baselines. However, the gap is closing rapidly. At the current doubling rate of roughly 79 days, the fleet could reach the insurance-claim benchmark (~300,000 MPI) within a few months if the trend holds.
The comparison is inherently imperfect because NHTSA's SGO 2021-01 reporting captures incidents at a lower severity threshold than police reports. Some robotaxi incidents are minor contacts that human drivers would never report. This makes the robotaxi MPI look worse relative to the police-reported human baseline, but the comparison to the "all crashes" estimate is more appropriate.
Why Simple Crash Rate Averages Are Misleading
Several news outlets have reported Tesla's robotaxi crash rate by dividing total incidents by total miles since launch. This yields a simple average that is significantly worse than the latest interval MPI. Headlines like "Tesla crashes 10x more than human drivers" are based on this calculation.
While mathematically correct, this approach is contextually misleading for a system that is rapidly improving. Here is why:
- Early performance dominates the average. The first few weeks of operation, when the fleet was small and the software was new, had the worst safety record. These early incidents drag down the cumulative average even though they are no longer representative.
- The trend matters more than the average. Judging Tesla by its cumulative average is like grading a student by their lifetime GPA instead of their most recent semester. If a student went from D grades to A grades, the lifetime GPA would be misleading.
- The latest interval is the best predictor. The most recent MPI values reflect the current software version and operational maturity. For safety assessment, the question should be "how safe is the system right now?" not "what was the average since day one?"
Our tracker displays both the interval-by-interval MPI and the exponential trend, giving a more complete picture than any single number can provide. For a deeper discussion of methodology, see our FAQ on methodology and data sources.
How Tesla Compares to Waymo
Waymo, the other major autonomous ride-hailing operator, reports approximately 1,000,000+ miles per incident based on their published safety data and the Swiss Re/Waymo study.
Tesla's current MPI of ~92,500 is roughly 10x lower than Waymo's. However, Waymo has been operating since 2009 and has accumulated billions of miles of testing data. Tesla launched its unsupervised service in June 2025. The rate of improvement matters: Tesla's safety is doubling approximately every 79 days, an unprecedented improvement trajectory in autonomous driving.
For a detailed comparison, see our Tesla vs Waymo safety analysis.
Where Does the Data Come From?
All incident data comes from NHTSA Standing General Order 2021-01, which requires manufacturers of Level 3-5 autonomous vehicles to report crashes to the federal government. Fleet size data is sourced from robotaxitracker.com, which monitors Tesla's API for active vehicles in Austin.
This tracker is fully open source. Our methodology is documented, our data is public, and our calculations are reproducible. For complete details, visit our FAQ page or read the methodology section on the main tracker.