Tesla’s fleet has accumulated over 1.2 billion miles on Autopilot and even more in ‘shadow mode’, report says

Authored by electrek.co and submitted by PrettyTarable
image for Tesla’s fleet has accumulated over 1.2 billion miles on Autopilot and even more in ‘shadow mode’, report says

Automakers and tech companies are rushing to bring autonomous driving systems to market and accumulating mileage with their test vehicles.

Tesla is taking a different approach by accumulating mileage with its customer fleet through its Autopilot driver assist program.

A new report now estimates that Tesla has accumulated over 1.2 billion miles on Autopilot and more than twice that when accounting for mileage in ‘shadow mode’.

The report was produced and is being updated by a MIT Human-Centered AI team led by Lex Fridman.

We reported on the team’s work with Tesla Autopilot in the past after they launched a study on driver interaction with Tesla Autopilot features last year by adding driver-facing cameras to Model S vehicles.

They released some interesting preliminary results and Fridman, the Research Scientist at MIT responsible for the study, said that he had been in contact with Tesla about his research and he believes the technology is required to enable more advanced autonomous features.

As part of their continuous work with Autopilot, they have been trying to track Tesla’s total miles accumulated using the driver assist system.

In a new report, they claim that Tesla has now accumulated over 1.2 billion Autopilot miles:

To be clear, that’s the mileage driven with Autopilot activated. Back in 2016, Tesla already had 1.3 billion miles driven with vehicles equipped with Autopilot. Tesla can still use data from those miles, but it’s not the same has mileage driven with Autopilot on.

That data is often referred to as “shadow mode” mileage because the autonomous driving system runs in the background of the vehicle without being able to have any input on the driving.

The MIT team has also tried to estimate the total of shadow mode miles:

In both cases, they plotted the data based on the few occasions when Tesla actually released some of the data.

It can be difficult to do that since the growth is exponential but not necessarily stable since Tesla said that it can detect dips in Autopilot usage by its fleet when there’s negative media coverage, which has been quite common over the last year.

What would be particularly interesting about Tesla crossing 1 billion miles of Autopilot use is that CEO Elon Musk has previously mentioned the milestone as the minimum they would need to move Autosteer from ‘beta’ to a regular feature.

Musk recently said that Autopilot is going to get some significant updates this summer and he says that progress could lead to the promised fully autonomous software being ready for the end of next year. Though he also said that Tesla’s version 9 software update is coming in August with the first ‘full self-driving features’.

As for the kind of data that Tesla collects from Autopilot, we recently got a rare great look at what Tesla Autopilot can see and interpret – pictured above.

mtreborv3 on December 3rd, 2018 at 00:55 UTC »

I'm curious how the 1.2 billion miles in a Tesla autopilot compar to 1.2 billion miles in a Tesla manually driven?

(Are Tesla drivers for luxury car drivers in general more careful with their cars than the average human driver?)

Edit: (in the same autopilot-allowable sections of road, e.g. On Freeways)

Hypothesis_Null on December 3rd, 2018 at 00:37 UTC »

This is a terribly stupid headline. Autopilot is only used in any significant amount in favorable weather conditions and very constrained driving conditions. Ie No rain or ice-covered roads and mostly highway driving, which is about the safest form of driving in terms of miles per fatality.

There is a reason that highway safety engineers spend so much time focusing on intersections. The weird double-diamond intersection that is cropping up is a result of trying to reduce 'critical points' where cars cross paths. These are responsible for the vast majority of car accidents, and last I checked, Teslas were not navigating any kind of intersection with their autopilots.

The only reasonable way to make a comparison is to take the subgroup of Drivers that live in Southern California. Then take the subgroup of that group that drive a car worth more than $35,000 to remove any sort of socioeconomic bias. Then measure their miles driven per crash incident - not fatality - and compare that against the rate of Tesla Drivers. That differential might be reasonably be attributed to the use of autopilot. You can't use fatalities because people are more likely to survive in a Tesla than in other cars.

Chances are there may be a small but notable difference. But none of this 4x nonsense. The only significant improvement I suspect autopilot will give is to correct mistakes of fatigued drivers. This might be undercut somewhat by people who drive fatigued because they know they have autopilot, though I would still expect a net-positive effect.

PM_ME_UR_QUINES on December 2nd, 2018 at 22:35 UTC »

I'd like to think that I already drive 4 times safer than the average person.