Independent Robotaxi Safety Tracking

When Will Tesla Robotaxis Operate
Without Safety Monitors at Large Scale?

Tesla has begun removing safety monitors from some Austin robotaxis. This tracker estimates when fleet-wide, large-scale removal becomes inevitable—using real-world incident data and safety trends, not announcements.

No hype. No insider claims. Just numbers.

Current Streak Live
--
miles since last incident
500K
Human driver benchmark (police-reported)

Robotaxi Safety Trend Analysis

Each data point shows the monthly Miles Per Incident (MPI) for Tesla Robotaxi, with an exponential trendline toward human driver crash rates

Miles per Incident
Miles Since Last Incident
Trend
Human - Police Reports (500K)
Human - Insurance Claims (300K)
Latest Interval
92,500
+61% from previous
Current Streak
--
miles since last incident
Total Incidents
10
Since June 2025
vs Human Drivers
5.4x
worse (latest interval)
Best Fit Model
Exponential
MPI = a·ebt
R² = 0.955
Doubling Time
79 days
Safety doubles every ~2 months
Daily Growth
+1.0%
Compound daily rate
30-Day Forecast
137,400
Projected miles/incident
Trend Projection

Based on the current exponential trend (R² = 0.955), Tesla Robotaxi MPI is projected to reach human-driver parity around --.

At a doubling time of ~79 days, the trend suggests MPI will cross the 500K police-reported crash benchmark by --. This projection assumes the current exponential improvement rate is sustained, which is not guaranteed.

How to Read This Chart

Understanding the safety trend visualization and what the numbers mean

Click to expand

Why Log Scale?

The Y-axis uses a logarithmic scale by default. This is standard for data spanning orders of magnitude (10K to 1M miles). On a log scale, equal vertical distances represent equal multiplicative changes (e.g., 10K→100K is the same distance as 100K→1M).

Straight Line on Log = Exponential Growth

A straight, upward-sloping line on a log chart means exponential improvement. Safety is doubling at a consistent rate. If the line curves upward, improvement is accelerating. If it flattens, growth is slowing.

What's Being Fit

The blue trend line fits an exponential curve (MPI = a·ebt) to the raw MPI data points (red dots) over time. Each red dot is the actual miles driven between two consecutive incidents. The fit is applied to MPI vs. time—not to cumulative data.

Linear vs. Log: No Tricks

Use the Log Scale / Linear Scale toggle above the chart to switch views. Linear shows absolute distances between values. Log reveals the growth rate. Both tell the same story—different emphasis. We default to log because the data naturally spans a wide range.

Fleet Size Over Time

Tesla Robotaxi fleet growth in Austin — unsupervised autonomous driving system (ADS) vehicles

Track the Expansion

See which cities Tesla is launching next and compare coverage with Waymo

View City Tracker

Tesla Robotaxi Safety Summary

Last updated:

As of January 2026, Tesla Robotaxi has driven -- miles without an incident in Austin, TX. The fleet of -- unsupervised Cybercab vehicles achieves a latest miles-per-incident rate of -- miles, with safety doubling every -- days. Compared to human drivers, who average 500,000 miles between police-reported crashes (NHTSA CRSS 2023), Tesla Robotaxi is currently --x less safe — but closing the gap rapidly. Waymo reports over 1,000,000 miles per incident, setting the industry benchmark for autonomous vehicle safety.

Top 5 Insights from Latest Robotaxi Incidents

Key takeaways from Tesla's autonomous vehicle safety data, updated with the latest NHTSA incident reports

# Insight Data Point Significance
1 Safety is improving exponentially MPI doubled from 14,100 to 92,500 in 6 months R² = 0.955 confirms strong exponential trend
2 Doubling time is consistent Safety doubles every ~79 days At this rate, Tesla could match human drivers in -- months
3 Current streak exceeds prior best -- miles since last incident Longest incident-free stretch since operations began June 2025
4 Fleet growing while safety improves Fleet grew from 10 to 72 vehicles Safety improvements hold even as fleet scales 7x
5 Gap with Waymo is narrowing Tesla at 92,500 vs Waymo at 1M+ MPI Tesla at ~--% of Waymo's safety level, up from 1.4%

Monthly Incident Summary

Monthly aggregated MPI estimates (NHTSA data has month-level resolution only) — see how we calculate MPI

Month Incidents Avg Fleet Total Miles MPI

Incident Composition Breakdown

Aggregated incident characteristics from NHTSA SGO reports — by movement type, speed, and collision partner

By Tesla Movement

Movement Type Count %
Stopped (0 mph) 4 24%
Proceeding Straight 9 53%
Backing (1-2 mph) 2 12%
Other/Unknown 2 12%

By Collision Partner

Partner Type Count %
Passenger Car 7 41%
SUV 3 18%
Fixed Object 3 18%
Cyclist/Pedestrian 2 12%
Bus/Other 2 12%

By Month

Month Incidents
Jul 2025 5
Sep 2025 4
Oct 2025 2
Nov 2025 1
Dec 2025 1
Jan 2026 4
17 Total incidents (Jul 2025 - Jan 2026)
100% Property damage only (no injuries)
0% Airbag deployments
100% Narratives redacted by Tesla
Tesla redacts the narrative field in all SGO reports, preventing determination of fault. Movement type, speed, and collision partner are from structured NHTSA SGO fields. Backing incidents (2 total) are low-speed parking lot events. Stationary incidents (4 total) occurred when the Tesla was at 0 mph.

Robotaxi Safety Comparison: Tesla vs Waymo vs Human Drivers

How robotaxi safety compares across Tesla, Waymo, and human driver crash rates

Tesla Robotaxi (Latest) 92,500
Tesla Robotaxi (Current Streak) --
Human - Insurance Claims 300,000
Human - Police Reports 500,000
Waymo (Reported) 1,000,000+

Understanding the Human Driver Benchmarks

Why we show two different comparison lines

500K - Police-Reported Crashes

~529,000 miles/crash

Source: NHTSA Crash Report Sampling System (CRSS) 2023

Calculation: ~6.14 million police-reported crashes ÷ 3,247 billion vehicle miles traveled = ~1.89 crashes per million miles

Limitation: Significantly undercounts actual crashes. NHTSA estimates 60% of property damage crashes and 32% of injury crashes are never reported to police.

This is the "official" statistic but represents a best-case scenario for human drivers.

300K - Insurance Claims

~307,000 miles/claim

Source: Swiss Re / Waymo study (2023), analyzing 500,000+ claims over 200 billion miles

Calculation: 3.26 property damage claims per million miles for human drivers

Advantage: Insurance claims capture incidents that go unreported to police, providing a more complete picture of actual crash frequency.

This is a more realistic comparison for AV incident data, which captures all crashes regardless of severity.

Why this matters: Tesla robotaxi incidents are reported to NHTSA under Standing General Order 2021-01, which requires reporting any crash where ADS was engaged within 30 seconds—including minor incidents that human drivers would never report. Comparing AV data to police-reported human crashes (500K) may overstate human safety. The insurance-adjusted benchmark (300K) provides a more apples-to-apples comparison.

Data Scope: Austin Only

Why all incidents are from Austin, TX

Austin, TX

Tracked

Unsupervised ADS (Level 4) - No human driver. Incidents must be reported to NHTSA under Standing General Order 2021-01.

Tesla has begun removing safety monitors from some vehicles in Austin. However, most of the fleet still operates with monitors present. This tracker estimates when large-scale removal becomes viable.

Bay Area, CA

Not Tracked

Supervised Testing - Human safety driver required by California law. Not subject to ADS incident reporting requirements.

164 vehicles operate with safety drivers per California regulations. Tesla must accumulate sufficient supervised miles before applying for driverless permits. Any incidents are reported under ADAS (Level 2) rules, not ADS.

All 10 Tesla incidents are from Austin — the only location with unsupervised robotaxi operations. Some Austin vehicles now operate without safety monitors, though most still have monitors. Bay Area vehicles operate with safety drivers as required by California law.

Fleet Size Math: Check the Numbers Yourself

The biggest uncertainty in MPI calculations is fleet size and utilization. Adjust the assumptions below to see how they affect implied fleet size and daily miles.

50 115 mi/day 250
4h 16 hrs 24h
10% 50% 90%
10 80 vehicles 200
Daily Fleet Miles 9,200
Monthly Fleet Miles 276,000
Implied Avg Speed 7.2 mph
Current Streak at These Assumptions --
Why this matters: Tesla does not publicly disclose exact daily miles per vehicle. Our default of 115 mi/day is derived from Tesla's Q3 2025 report (~250K total miles). Fleet size comes from robotaxitracker.com. Changing these assumptions shifts all MPI values proportionally—the trend direction stays the same, but absolute values change. This is the primary source of uncertainty.

Methodology

How we calculate miles per incident for the safety trend analysis

01

Fleet Data

We use total fleet size (all vehicles in the Austin fleet) scraped from robotaxitracker.com, interpolated daily between known data points.

02

Miles Estimation

Daily miles = Fleet size × 115 mi/vehicle/day (based on Tesla's 250K miles Q3 2025 report).

03

Incident Data

Incidents from NHTSA Standing General Order reports. NHTSA data has monthly resolution only (e.g., "JAN-2026"), so we aggregate incidents by month and show monthly MPI estimates. Tesla has been cited for delayed reporting.

04

Trend Analysis

Exponential model (MPI = a·ebt) fitted via log-linear regression on monthly aggregate data. R² computed in log scale, ensuring 0-1 bounds.

How This Tracker Estimates Large-Scale Monitor Removal

Leading indicators for when monitor-free robotaxi operation at scale becomes inevitable

Tesla has already begun removing safety monitors from some vehicles in Austin, but the majority of the fleet still operates with monitors. In the Bay Area, safety drivers are required by California law, and Tesla must accumulate sufficient supervised miles before it can apply for driverless permits.

Large-scale monitor removal depends on multiple thresholds being crossed simultaneously. This tracker focuses on leading indicators, including:

  • Miles per incident approaching or falling below human benchmarks
  • Sustained operation across multiple cities without regression
  • Stability of deployed software versions over weeks rather than days
  • Accumulation of sufficient supervised miles in new markets (e.g., Bay Area) to satisfy regulatory requirements

When these signals align, the probability of large-scale safety-monitor removal rises sharply, even before any official announcement.

This approach does not answer the question "Has Tesla removed safety monitors yet?" (it already has, in some Austin vehicles). It answers a different question: "When will removal happen at large scale, across the fleet?"

Robotaxi Safety vs. Public Opinion on X

Data-backed responses to common claims about robotaxi safety on social media

Common Claim

"Tesla robotaxis crash all the time"

What the Data Shows

Tesla Robotaxi has had 10 incidents in over 7 months of operation across a fleet of 72 vehicles. The latest interval between incidents is 92,500 miles — and the current streak without any incident is -- miles. By comparison, human drivers in the U.S. average a crash every 300,000–500,000 miles (NHTSA CRSS).

Common Claim

"Robotaxis can't handle rain or bad weather"

Addressing Rain Disengagement Myths

Tesla's Austin fleet operates in a region that receives approximately 34 inches of rain annually. The fleet has continued operations through multiple weather events. During the January 2026 ice storm, operations were paused for 2 days — the same period when Austin's public transit (CapMetro) and city services were also suspended. Weather pauses are a safety feature, not a failure mode.

Common Claim

"Tesla is hiding crash data"

What the Data Shows

All Tesla Robotaxi incidents are reported to NHTSA under Standing General Order 2021-01, which requires reporting any crash within 30 seconds of ADS engagement — including minor incidents that human drivers would never report. Tesla has been cited by NHTSA for delayed reporting, but all incidents are eventually publicly disclosed. This tracker independently monitors this data.

Common Claim

"Waymo is much safer — Tesla should give up"

What the Data Shows

Waymo does report higher MPI (~1M+ miles per incident), but Tesla's safety is doubling every ~79 days. Waymo has been operating since 2009; Tesla's unsupervised robotaxi launched in June 2025. Tesla's rapid improvement trajectory could close the gap if the exponential trend (R² = 0.955) continues. Both companies are making roads safer.

Download the Dataset

Open data for researchers, journalists, and developers. Use it in your own analysis, cite it in articles, or feed it to LLMs.

Dataset Fields (Monthly Aggregates)

month Month and year (e.g., "Jul 2025")
incident_count Number of incidents in the month
avg_fleet_size Average fleet size during the month
total_miles Estimated fleet miles for the month
mpi Miles per incident (total_miles / incident_count)

Note: NHTSA SGO data has month-level date resolution only, so incidents are aggregated by month.

Sources: NHTSA SGO 2021-01, robotaxitracker.com, Tesla Q3 2025 report. Licensed under MIT.

About the Creator

Independent analysis by Kangning Huang

This robotaxi safety tracker was created to provide transparent, data-driven insights into Tesla Cybercab safety performance. Follow for more analysis on robotaxi safety, autonomous vehicles, and technology.

Robotaxi Safety FAQ

Common questions about robotaxi safety, Tesla Cybercab crash rates, incident data, MPI methodology, and what the headlines get wrong

Tesla has already begun removing safety monitors from some vehicles in Austin, but this is not yet at large scale. Fleet-wide removal is expected only after safety metrics consistently match or exceed human drivers across cities and scale. In the Bay Area, California law requires safety drivers, and Tesla must accumulate sufficient supervised miles before applying for driverless permits.

Key conditions for large-scale monitor removal include sustained MPI at or above human-driver benchmarks (300,000–500,000 miles), geographic expansion without incident regression, stable software versions across the fleet, and meeting regulatory requirements in states like California.

Fleet growth rate, intervention density, geographic expansion, and software stability are more predictive than press releases or earnings calls.

This tracker focuses on these quantitative signals because they reflect operational reality — the conditions under which removing safety monitors at large scale becomes economically rational, regardless of what has been publicly announced.

Robotaxi safety is primarily measured using miles per incident (MPI) — the average distance a robotaxi drives between crashes or reportable incidents. Higher MPI means better safety. In the U.S., all autonomous driving system (ADS) incidents must be reported to NHTSA under Standing General Order 2021-01, providing a transparent public record of robotaxi safety performance.

This tracker monitors robotaxi safety for Tesla's Cybercab fleet in Austin, TX — the only fully unsupervised (Level 4) Tesla robotaxi deployment. Key robotaxi safety metrics include MPI trend over time, comparison to human driver crash rates, and fleet-wide incident frequency. Waymo, the other major U.S. robotaxi operator, publishes its own safety data showing over 1,000,000 MPI.

Tesla robotaxis in Austin average approximately -- miles between incidents. For comparison, human drivers average about 500,000 miles between police-reported crashes, or approximately 300,000 miles between insurance claims.

Tesla's safety is improving rapidly, with the interval between incidents doubling approximately every -- days.

Tesla has reported -- incidents to NHTSA since launching in Austin in June 2025. All incidents occurred in Austin, Texas, which is the only location where Tesla operates true unsupervised autonomous driving (Level 4).

The Bay Area fleet operates with safety drivers and follows different reporting requirements.

Yes, significantly. Our analysis shows Tesla robotaxi safety is doubling approximately every -- days. The most recent interval between incidents reached -- miles — a --% improvement from the previous interval.

This exponential improvement trend (R² = --) suggests the system is learning and becoming safer over time.

Waymo reports approximately 1,000,000+ miles per incident based on their published safety data. Tesla's latest interval is -- miles per incident — roughly --x less than Waymo's reported figures.

However, Tesla's rapid improvement rate (doubling every ~-- days) could narrow this gap significantly if the trend continues.

Tesla currently operates robotaxis in two markets: Austin, Texas (unsupervised, with some vehicles already running without safety monitors) and the San Francisco Bay Area (with safety drivers, required by California law). Tesla must accumulate sufficient supervised miles in California before applying for driverless permits.

Tesla has announced plans to expand to Dallas, Houston, Phoenix, Miami, Orlando, Tampa, and Las Vegas in the first half of 2026.

As of the Q4 2025 earnings call, Tesla reported "well over 500" vehicles across Austin and the Bay Area carrying paid customers. Elon Musk stated the fleet is expected to "double every month."

Our tracker monitors the Austin fleet specifically, as it's the only location with true unsupervised autonomous driving.

We calculate miles per incident by:

  • Tracking fleet size daily from robotaxitracker.com and news sources
  • Estimating daily miles using 115 miles/vehicle/day based on Tesla's Q3 2025 report
  • Recording incidents from NHTSA Standing General Order crash reports

We then fit an exponential trend model to identify the improvement rate.

Austin is the only location where Tesla operates unsupervised Level 4 autonomous driving, which requires incident reporting to NHTSA under Standing General Order 2021-01. Some Austin vehicles already operate without safety monitors, though most still have monitors present.

The Bay Area fleet operates with safety drivers (Level 2) as required by California law — Tesla must accumulate sufficient supervised miles before applying for driverless permits. Different reporting requirements apply, so comparing only Austin data ensures consistency.

As of January 2026, the current Tesla Robotaxi Miles Per Incident (MPI) is -- miles. This means the Tesla Cybercab fleet in Austin, TX drives an average of -- miles between each reported incident.

  • Latest MPI: -- miles per incident
  • Improvement rate: Doubling every -- days
  • 30-day forecast: -- miles per incident
  • Total incidents tracked: -- since June 2025
  • Current streak: -- miles without an incident

MPI is calculated by dividing total fleet miles driven by the number of incidents in each interval, based on NHTSA Standing General Order data.

As of January 2026, Tesla Robotaxi safety compared to human drivers:

  1. vs. Police-reported crashes (500,000 MPI): Tesla is --x worse, but improving exponentially
  2. vs. Insurance claims (300,000 MPI): Tesla is --x worse, a more realistic comparison since Tesla reports all incidents
  3. vs. Waymo (1,000,000+ MPI): Tesla is --x worse than Waymo's reported figures

Key context: Tesla's safety is doubling every ~-- days. At this rate, Tesla could match human driver safety levels (insurance-adjusted) within months if the exponential trend continues. Data sourced from NHTSA CRSS 2023 and the Swiss Re/Waymo safety study.

As of January 29, 2026, Tesla Robotaxi has driven -- miles without an incident in Austin, TX. The fleet of -- unsupervised vehicles has been operating incident-free since --.

This ongoing streak already exceeds the latest completed interval of -- miles, suggesting continued safety improvement.

Tesla Robotaxi crash frequency as of January 2026:

  • Total incidents: -- reported to NHTSA since June 2025
  • Latest crash rate: 1 incident per -- miles driven
  • Improving trend: Crash frequency is decreasing, with MPI doubling every -- days
  • Data source: All incidents reported under NHTSA SGO 2021-01, which requires reporting any crash within 30 seconds of ADS engagement

Note: This crash frequency captures all incidents regardless of severity, including minor ones that human drivers would typically never report to police.

Several news outlets have reported alarming crash rate comparisons:

  • Carscoops: "Tesla's Robotaxi Crash Rate Is Way Worse Than We First Thought"
  • Common Dreams: "Tesla Robotaxis Are Crashing More Than 12 Times as Frequently as Human Drivers"
  • Electrek: "Tesla's own Robotaxi data confirms crash rate 3x worse than humans"
  • Electrek: "Tesla Robotaxi had 3 more crashes, now 7 total"
  • Futurism: "Tesla's Robotaxi Crashes Four Times In a Single Month"

These reports rely on simple averages of past performance — dividing total incidents by total fleet miles since launch. While mathematically correct, this approach misses what the data actually shows: a rapidly improving safety trend.

What the trend analysis reveals:

  • Current incident-free streak: -- miles without an incident — already approaching the human driver benchmark of 500,000 miles between police-reported crashes
  • Safety is doubling every -- days (R² = --), an exponential improvement that simple averages completely hide
  • Latest completed interval: -- miles per incident — already --x better than the simple average of ~-- MPI across all -- incidents

The news articles treat the crash data as a static snapshot. Our analysis tracks the trajectory — and the trajectory shows a system that started poorly but is improving at an exponential rate. The current streak, with a fleet of -- vehicles driving daily, already demonstrates near-human-level safety performance.

Additionally, NHTSA SGO 2021-01 requires reporting all incidents — including minor ones that human drivers would never report to police. Human crash statistics (1 per 500,000 miles) only count police-reported crashes, making direct comparisons inherently unfair.

Bottom line: Judging Tesla Robotaxi's current safety by a simple average of all past performance is like judging a student's current ability by their lifetime GPA instead of their latest semester. The data shows a system that is learning and improving — and the current streak already rivals human driver safety levels.

Yes. Multiple news outlets have reported that Tesla redacts crash narrative details from NHTSA Standing General Order reports:

  • Carscoops reported that Tesla redacts the "narrative" section of each crash report, preventing the public from knowing how crashes happened
  • Sherwood News noted that one crash was more heavily redacted than usual, with no disclosure of what the robotaxi hit
  • WebProNews highlighted the contrast with Waymo's more transparent reporting

While this is a legitimate transparency concern, it does not affect the core safety metrics tracked here. This tracker uses only the structured, non-redacted NHTSA SGO data fields:

  • Incident dates — reported to NHTSA (not redacted)
  • ADS engagement status — whether autonomous driving was active (not redacted)
  • Fleet size — tracked daily from robotaxitracker.com
  • Daily miles — estimated at 115 mi/vehicle/day from Tesla's Q3 2025 report

The MPI trend (R² = --) and safety doubling time (-- days) are fully calculable from non-redacted data. More narrative transparency from Tesla would be welcome, but the quantitative safety trajectory is clear from the data we do have.

An "incident" on this site means a crash event that appears in NHTSA Standing General Order (SGO) 2021-01 ADS crash reports for Tesla's Austin robotaxi program. This can include minor low-speed contact that human drivers often wouldn't report to police.

Incidents may be not-at-fault (e.g., being rear-ended). We do not count near-misses or disengagements unless they result in a reportable crash.

Not yet, on a simple average basis. Tesla Robotaxi's latest interval between incidents is -- miles, compared to human drivers' 500,000 miles between police-reported crashes. However, safety is doubling every ~-- days, meaning Tesla is on an exponential trajectory toward parity.

The comparison is also complicated by reporting differences: NHTSA SGO 2021-01 captures all incidents (including minor ones), while human crash statistics only count police-reported events. Using the insurance-adjusted benchmark (300,000 miles), the gap is narrower.

Miles per incident (MPI) measures the average distance an autonomous vehicle drives between reportable incidents. It is the primary safety metric for comparing AV performance against human drivers and other AV operators like Waymo.

Higher MPI = safer vehicle. A human driver averages ~500,000 miles between police-reported crashes (~300,000 between insurance claims). Waymo reports 1,000,000+ MPI. MPI matters because it normalizes safety across fleets of different sizes and operating hours.

Yes. NHTSA Standing General Order 2021-01 requires AV operators to report any crash where the automated driving system was engaged within 30 seconds of the incident—regardless of severity. This includes minor fender-benders, parking lot scrapes, and incidents where the AV was stationary and struck by another vehicle.

By contrast, human crash statistics (500,000 MPI) only count police-reported crashes, which miss an estimated 60% of property-damage-only incidents. This reporting asymmetry makes direct AV-to-human comparisons inherently skewed against AVs.

NHTSA SGO 2021-01 requires initial crash reports within 1 day and updated reports within 10 days. However, Tesla has been cited for delayed reporting. This means the "current streak" number may be artificially high if an incident occurred recently but hasn't yet appeared in the public NHTSA database.

We treat streak data as provisional until the reporting window catches up (typically 2-4 weeks). Historical MPI intervals for completed periods are more reliable than the ongoing streak figure.

Robotaxi incidents are reported under NHTSA SGO 2021-01, which can capture more minor events than police-reported human crash stats. Robotaxis also operate in a limited geofenced ODD, while national human benchmarks include highway/rural driving.

Finally, SGO crash reports can arrive with reporting lag, so "current streak" numbers near the latest date are best treated as provisional until the reporting window catches up.