Tracking Autonomous Miles: Why is it So Important?

//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>

AV data is growing and evolving rapidly, and understanding the significance behind the statistics is essential to fully grasping the market potential of AV companies. There are at least two categories of data that are crucial for understanding this progress, though autonomous vehicle miles travelled (VMT) is the most common measure for indicating an AV company’s experience and safety thanks not only to the sheer volume of data available, but also the historical length of the data itself.

The usage of several growth measures for VMT, however, can make it difficult to compare the progress of different AV companies. With this in mind, we will analyze autonomous VMT growth measures that are used, their definition and how they provide useful information—more specifically in terms of robotaxis deployments, as there is more data for this AV segment compared to other use cases.

Counting autonomous VMT

As robotaxi deployment grows, statistics on the number of paid trips and related data become important. Today, there is limited public data, with Waymo providing a majority of the available information used to formulate recent growth perspectives.

In most instances, the number of autonomous passenger trips is used to illustrate progress, such as in the case of AVs goods delivery (i.e., the number of cargo loads delivered by robotrucks). For example, the growing volume of autonomous VMT is indicative of testing efforts and experience with developing AV software driving platforms that lead to better AV safety. As AV deployment begins in earnest, cumulative growth of VMT into the tens of millions of miles, with data revealing no or few severe crashes compared to human drivers, is a potential indicator of increased AV safety.

The next table looks at the categories of AV miles that are commonly used. The first entry is AV driving simulations or virtual VMT, which are often used to train software driver platforms and is considered the most important technology for developing safe AV software (despite its lack of attention in the trade press). Every major AV developer has made large efforts in AV simulation to improve their AV software driver safety.

For instance, Waymo is a leader in virtual VMT testing and simulation. In April 2020, Waymo recorded 20 million miles per day in AV simulation, with cumulative simulation VMT surpassing 15 billion miles.

Self-driving miles is usually the first measure of AV experience used to show a company’s autonomous experience. This category normally includes VMT with safety drivers (referring to experienced human drivers who supervise an AV during its testing phase) onboard as most regulations and permits require a safety driver as the first step in AV testing.

Driverless VMT is the next logical step in reporting AV miles. This stage usually focusses on driverless autonomous VMT without safety drivers. Testing starts with AV company employees onboard, then progresses to providing transportation to employee travel—mostly work-related trips.

Passenger VMT refers to rides with passengers, which begins with limited groups and expands as VMT data provides more insight into experience and safety. This testing stage may utilize a safety driver when driverless permits are restricted. The instance of passenger VMT where safety drivers are present is mostly used in China, but limited in the U.S. Driverless passenger VMT is the preferred method to gather AV data in the U.S.

Paid passenger VMT are another avenue of data retrieval and may follow a similar path as above, including safety driver use as a first step in testing in China, with driverless testing used at a later stage.

The rider-only designation is primarily used by Waymo, and the company only counts VMT when passengers are transported. Waymo’s use is almost exclusively paid trips with customers. A key characteristic is that miles between rider-only trips are excluded from the VMT statistics. Waymo cannot charge revenue for these “empty” VMT trips. Other AV companies are likely to use this classification as they expand deployment.

The next figure is a pictorial view of current VMT statistic usage with deployment segments on the y-axis and unspecified time on the x-axis.

The top green box shows information on AV simulation activities that help improve software driver safety and reliability. The safety efforts are focused on understanding and learning how to manage previously unknown edge cases or unfamiliar traffic events. Simulating recorded crashes is also of high value.

The bottom left blocks show two stages of self-driving with safety drivers. The second of the lower-left black blocks shows early passengers, which includes employees and later small groups of public users. The third black block shows paid passengers in self-driving AVs with safety drivers (this is usually a small part of deployment and is used before regulation allows driverless activities.

The first of the two blue blocks show driverless VMT activities with no passengers, eventually transitioning to employees and free trips for small groups of public users. The third blue block shows driverless AVs as they expand to new cities and repeat the pattern of the first two blue blocks.

The fourth blue block will eventually become the largest self-driving VMT category, which is empty VMT trips between rider-only trips. It is challenging to estimate the empty driverless rides without public data. The California Public Utilities Commission (CPUC) is in the process of collecting this data.

The first two red blocks show rider-only trips, which begin with employees and soon move to free rides with small consumer groups. The paid rider-only category is the next red block and is the expansion of rider-only deployment in specific cities.

The third red block represents new city expansion, which will have the same pattern as the first two red blocks. Waymo has since added four new cities to its testing pool, with at least six major cities in the pipeline. Five cities in Silicon Valley are in the employee/small consumer stage and close to paid deployment.

Historical Waymo autonomous VMT

Waymo has by far the most autonomous VMT in all categories described above and has the most historical data. Fortunately, Waymo also has the most publicly available data. Waymo has provided VMT data in three categories over the last decade:

• Self-driving VMT with safety drivers was the first measure used by Waymo to collect public data starting in 2015.
• Driverless VMT without safety drivers was the second category Waymo use to collect public data starting in 2019 once regulation allowed driverless AV use cases.
• Waymo’s use of rider-only VMT is the third measure used by the company to collect public data and is quickly becoming its most important measure of deployment success for any city and region.

Rider-only progress is typically expressed as cumulative VMT for a specific city, region or AV company. Cumulative values for a specific period are needed to compare AV crash statistics with human driver crash rates. Since December 2022, Waymo has released seven instances of rider-only VMT data categorized by cumulation of miles by cities.

Waymo is using rider-only VMT and its crash statistics to illustrate that its driverless software is safer compared to human drivers. The latest Waymo publication, which reports on data up to June 2025, compared its crash data after its rider-only VMT reached nearly 96 million miles.

The next figure is a compilation of Waymo’s public VMT data, with added estimates for year-end 2025, 2026 and 2027. The top green box is public data on Waymo’s simulation AV miles—both daily and cumulative values. Daily AV simulation miles grew from three million in 2016 to 20 million in 2020. With no public data after 2020, I have assumed 20 million miles per day from 2021 through 2027. This is on the low side, but shows how important the volume simulation of VMT is.

The cumulative simulation miles are measured in billions and grew from 3.5 billion in 2018 to over 20 billion in 2020. Assuming 20 million simulation miles per day increases cumulative miles by over 7 billion miles per year, the estimated cumulative miles by year-end 2025 is over 55 billion.

The self-driving miles with safety drivers are shown in the bottom black blocks. Waymo has collected a lot of public data through 2020 but has provided limited self-driving data since then.Waymo’s robotaxi testing, however, shifted to driverless VMT in 2021 and rider-only miles after 2022. I think self-driving with safety drivers will only be used in the early stages of new city testing. This includes exploring the potential of new cities with different driving cultures, which Waymo refers to as “road trips.” Hence, self-driving VMT will increase slowly from 25 million in October 2022 to around 40 million VMT in 2027.

The blue blocks show Waymo’s driverless VMT trends. There are no safety drivers in Waymo’s driverless VMT figures. Driverless VMT reached one million miles at year-end 2019 and grew to 20 million in 2023.

The driverless category is now benefitting from Waymo’s empty miles between rider-only trips because paid passenger trips are now on a rapid deployment expansion. This category also includes the time it takes to charge, clean and execute other overhead driverless activities. It is likely that empty VMT as a ratio to rider-only VMT will decline as customer usage rates and operational efficiency increase. Waymo’s driverless VMT is estimated to reach 70 million at end of 2025. Future growth will be strong as empty and overhead miles will be proportional to rider-only growth.

The red blocks summarize Waymo’s rider-only VMT since December 2022. The top red block shows 96 million for June 2025, which is projected to reach 165 million VMT at year-end 2025. With Waymo planning to add at least five more cities in 2026 and expand testing in existing cities, the year-end 2026 VMT is forecasted at 350 million rider-only cumulative miles—a growth of over 110%. VMT growth in 2027 is estimated at nearly 100% and could reach 690 million cumulative VMT or 340 million VMT for calendar year 2027.

AV VMT and crash rates

Large numbers of driverless and rider-only VMT with few crashes are crucial to verify that robotaxi software drivers are safer than human drivers. The most important statistics to back this claim are that there are no Waymo fatalities for at least a few more years. Currently, fatality crash rates for human drivers in the U.S. are about 1.2 deaths per 100 million VMT.

However, this includes fatality crashes from alcohol-impaired drivers, speeding and un-belted drivers/passengers. Robotaxi operation can prevent these risky human driver fatalities. Hence, it would be best if robotaxi fatalities are compared to human fatalities without DUI, speeding and un-belted fatalities included.

Human fatalities from risky driving account for about 44% of total yearly U.S. crash fatalities, which may vary slightly by year. Some fatalities have two or three of these factors as crash contribution. This overlap has been eliminated in the 44% figure. Risky driving behavior includes distraction factors like smartphone usage as a significant crash contributor. However, crash statistics on smartphone contribution are limited.

I will assume 40% of crashes are due to risky human behavior and compare robotaxi drivers with cautious/lawful human drivers. This implies that the fatality crash rate for cautious human drivers is about one fatality per 200 million VMT. Robotaxi crash rates need to be at least as good as this figure.

This means each robotaxi company may need to drive at least 200 million miles without an AV-caused fatality to have some confidence that their software is safer than cautious human drivers. Even higher VMT may be needed for AV assurance of better safety than human drivers.

Waymo is furthest along with reaching 200 million autonomous VMT. If we include driverless and rider-only VMT, Waymo has topped 150 million miles and will be close to 200 million by year-end 2025. By end of 2026, Waymo will probably top 450 million VMT if driverless miles are included and could reach 900 million miles by year-end 2027.

It is notable that Waymo is more conservative and exclusively counts its rider-only VMT in its crash comparison with human drivers. But Waymo is comparing its crash rates to average human drivers, not the cautious human driver. Additionally, Waymo uses localized crash rates that match the zip codes of Waymo’s crashes, and this data may not be available for cautious drivers.