Last week, Tesla CEO Elon Musk made an interesting comment that went mostly unnoticed during a talk in Norway. While talking with Minister of Transport and Communications Ketil Solvik-Olsen, Musk said that based on early data from the Autopilot program, Tesla’s semi-autonomous driving system, the probability of an accident happening is about 50% lower when the system is activated.
Here are Musk’s exact words:
“The probability of having an accident is 50% lower if you have Autopilot on. Even with our first version. So we can see basically what’s the average number of kilometers to an accident – accident defined by airbag deployment. Even with this early version, it’s almost twice as good as a person.”
He is referring to the first generation of the Autopilot as he expects the second generation, which should be fully autonomous, to be significantly safer as it will need to be in order to be made legal by regulators.
“I think it’s going to be important in term of satisfying regulators and the public to show statistically with a large amount of data – with billions of kilometers of driving – to say that the safety level is definitively better, by a meaningful margin, if it’s autonomous versus non-autonomous.”
While Tesla is far from getting to billions of kilometers of data, the automaker confirmed earlier this month that Tesla owners have already driven over 47 million miles (75 million km) on Autopilot since officially launching the feature in October 2015. The data is increasing exponentially as Tesla’s fleet is growing by tens of thousands of vehicles every quarter.
Musk expects Tesla’s fully autonomous Autopilot system to be ready within the next two years. While the technology should be ready by then, Musk doesn’t know exactly how long the regulatory process will take, but he is confident the data should convince them, especially since the early data from the first generation Autopilot is encouraging.
Featured Image: A member of the media test drives a Model S equipped with Autopilot in Palo Alto, California, U.S., on Wednesday, Oct. 14, 2015. Photographer: David Paul Morris/Bloomberg via Getty Images
FTC: We use income earning auto affiliate links. More.
I would want to see the raw data. He defines an accident as when the airbag is deployed. I’ve been in several things that I would call accidents and I’ve never had an airbag deployed.
I’d be concerned that, when including low speed collisions, maybe autopilot makes those more likely?
Well there’s also bias as to when autopilot gets used. If they’re comparing statistics of autopilot use vs all other driving, then they are effectively comparing specifically highway driving to all other driving.
It’s also possible that he’s talking about stats across the whole fleet in autopilot-equipped cars, whether or not they have the auto-steering enabled. That would include improvements from emergency braking and side impact avoidance even if the driver is controlling the vehicle.
Here is your data…
https://www.youtube.com/watch?v=9I5rraWJq6E
That’s not data. That’s an anecdote.
Maybe if he didn’t have autopilot he would have been paying more attention and seen that guy was trying to make exit and let him in. It happens to me all the time and I slow down in anticipation and let them in.
Totally unscientific data he spews out. As other point there is no control so he is comparing all driving to only when people are comfortable to put autopilot.
Also, most cars now have alarms or sensors that help alert when something like that is happening he is comparing to using nothing. Its like comparing a new antibiotic against using nothing at all.. its has little value unless you compare to the current antibiotics or treatment technology. Finally this is ridiculous because autopilot will undoubtedly be inferior because obviously its clear no car manufacturer can avoid issues with electronic things working pefectly with airbags, to ignition switches to doors even so when people start relying completely on a electronic system working you open a whole new can of worms also so the benefit of the technology is diminished.
My reading of his statement is that cars that regularly use auto-pilot typically have 50% more miles between accidents than cars that don’t.
I suspect this is misleading for a number of reasons, not the least of which is that auto-pilot implies highway driving, which is inherently much safer than city driving. So, it doesn’t sound like an apples-for-apples comparison. Also, air bag deployment is a fairly serious accident.
I’m all for auto-pilot and believe all cars will eventually be mandated to have it, but I also believe (like Don Norman) that Tesla’s been fortunate to this point that there hasn’t been a catastrophic accident caused by a distracted driver who got too comfortable with something that isn’t truly 100% autonomous. It’s seemingly an inevitable outcome, and would create a firestorm of negative publicity.
I read it as “50% more highway miles driven between accidents when autopilot is engaged on the Model S than when it isn’t engaged”.
The majority of highway accidents occur at on/off ramps and intersections, where a driver is less likely to be using autopilot. The data could suggest that autopilot is more likely to be used in less accident prone scenarios in the first place, not that autopilot reduces the amount of accidents, as much as it could suggest what Musk says.
That’s the way it’s worded, and was my first impression as well — however, that *implies* that people are crashing with autopilot ENGAGED. And not just getting into a 2MPH sideswipe as a Google autopiloted car did (which was front page news) but crashing hard enough to deploy airbags…
All in all, that makes it extremely unlikely to be the wording that Musk intended.
Autopilot crashes are front page news. And a high-speed autopilot crash has never happened as far as I’m aware.
So, I suspect they’re looking at autopilot-enabled cars vs non, and the subset of users who regularly use it, and pulling their statistics from that.
I agree with your points. I’d also like to know the raw count of how often airbags have deployed versus what would be expected. I, like most people, don’t have any intuitive sense of the statistical significance of “50% fewer” airbag deployments in 47 million miles traveled. (i.e., Are we talking about 2 vs. 4? 25 vs. 50?)
ok.
180 days between AP release and Elon reporting 47mil miles. thats is ~270,000 miles a day. we know more Teslas are being added to that number daily. Also, w/ winter over, more people will be driving with AP. In the 2 weeks between then and know, AP should have gained another ~3.3-3.7 million miles, bringing the total to over 50 million miles of Autopiloted Driving. So expect another 50-60 million miles before Auld Laud Sang 2017. They will have plenty of data to present to legislatures I’m sure.
270,000 miles per day? There are only 100,000 Model S’ out there which would imply that each and every one of them has been updated with Autopilot and drive 270 miles per day, every day. Some don’t even have that kind of range.
Just to let you know, calculation in your post doesn’t work… (270000/100000=2.7 not 270…)
Damn it, you’re right. It was late last night. My whole point is moot then.
“As of December 2015, a total of 107,228 Model S sedans have been sold worldwide since its introduction, ranking as the world’s second best selling plug-in car ever after the Nissan Leaf (200,000)”
“Model S global sales totaled 12,420 units during the first quarter of 2016.” source wikipedia.
there are over 120,000 MS floating around by yesterday.
I use it daily and it’s really good. Still has areas when you won’t want to use it. That’s usuay on ramps to go on another freeway. Sometimes it still has difficulties with reading lines and can get weird. Also I have phantom braking and not sure why. Happens in the same spot every day. Not a dramatic braking but it does brake. All in all though I use it everyday in carpool. It’s great and will continue to get better. One draw back, you can often forget if it’s driving even with the blue steering wheel lit.
Can someone more knowledgeable on the topic answer, whether Tesla can use the driver input data to train autopilot even when autopilot is turned off?
The driver input data is sent while in autopilot mode and the driver intervenes with what the autopilot wants to do. An example would be when the car mistakenly wants to follow the lane markers off of a freeway exit.
As a former Ford Motor Company employee I an amazed that a car company can actually be honest and fair to the public and have not once seen a trash talk of its fierce compitetion. Tesla will eventually be the leader in auto industry and hopefully able to continue with providing value to its buyers.
Great job
I think he’s referring to Automatic Emergency Braking reducing “accidents”. But ramming the car in front of you because you are texting is not an accident, it’s a crash. Accidents are unexpected outcomes, but hitting something with a moving car because you are not controlling is not unexpected, it’s a predictable outcome.
Teslas are one of the more expensive cars to insure in it’s price range.
How can autonomous cars not be safer? They’re reactions are literally fast as light and always vigilant. Soon the majority of cars will be talking to each other and know where they are and what their intentions are, improving their safety even more.
75 million km is like driving from Earth to Mars.
It’s a great strategy to get out first, provide all this road driven data to regulators. Maybe regulators will make this requirement before approval.