Is Autopilot safer than a human driver in Great Britain?

[NewbieT]

Elon recently celebrated Autopilot’s accident record claiming it had a nearly 10 times lower chance of having an accident.

https://twitter.com/x/status/1383548841438236674

Tesla Vehicle Safety Report


So, is Autopilot better than a human in Great Britain?

In 2019:

• 356.5 Billion Vehicle Miles for all motor vehicles [Road traffic statistics - Summary statistics, Vehicle Miles Driven, All Motor Vehicles]
• 117,536 accidents [Reported road accidents (RAS10)]

One accident for every 3.03 million miles driven.

Now if we just look at Motorways (the current intended use case for Autopilot):

• 70.5 Billion Vehicle Miles
• 4,130 accidents

One accident for every 17 million miles driven.

I don’t think we need to start protesting to ban human drivers just yet.

Tesla’s methodology doesn’t look at minor accidents. As a comparison there are about 2.1m motor claims in the UK each year = One per 13,000 or so miles. The 117k figure is accidents recorded by the police. The long and short of it because Tesla doesn’t produce their own regional breakdown it’s hard to say. Depending on how you cut the accident data you can make the numbers say what you like. I would suggest that perhaps the NHTSA comparison is not a fair one.

Interestingly, it does seem the UK is much safer place to drive than the USA generally though.

International comparisons of road accidents (RAS52)

 

[Alan J]

Other than Phantom Breaking which is an occasional but scary occurrence I would say yes it is.

 

[M3noob]

Other than Phantom Breaking which is an occasional but scary occurrence I would say yes it is.

Braking is the problem. My car isn't broken. And much less of a problem when on the continent. Possibly a RHD issue?

 

[Wol747]

I would be very interested in way the statistics are presented.
Does Tesla say whether the accident cars were being driven in autopilot when the crashes actually occurred? It wouldn't be a lie to say that a crash happened a couple of seconds after the driver had had to disconnect because the autopilot was doing something wrong and therefore it was a "human" crash and not an autopilot one but it would certainly be manipulation of the statistics.
I would like confirmation of exactly what "autopilot engaged" actually means before accepting the figures.

 

[NewbieT]

Methodology:
We collect the exact amount of miles traveled by each vehicle with Autopilot active or in manual driving, and do so without identifying specific vehicles to protect privacy. We also receive a crash alert anytime there is a crash that is correlated to the exact vehicle state at the time. This is not from a sampled data set, but rather this is exact summations. To ensure our statistics are conservative, we count any crash in which Autopilot was deactivated within 5 seconds before a crash, and we count all crashes in which the crash alert indicated an airbag or other active restraint deployed. In practice, this correlates to nearly any crash at about 12 mph (20 kph) or above, depending on the crash forces generated. On the other hand, police-reported crashes from government databases are notoriously under-reported, by some estimates as much as 50%, in large part because most fender benders are not investigated. We also do not differentiate based on the type of crash or fault, and in fact, more than 35% of all Autopilot crashes occur when the Tesla vehicle is rear-ended by another vehicle. In this way, we are confident that the statistics we share unquestionably show the benefits of Autopilot.

They’ve been transparent in their methodology. Ignoring the x10 reference from Elon they’re saying Autopilot it about twice as safe. It’s just not a like for like comparison when you’re comparing ‘mostly motorway’ Autopilot with ‘all roads’ human driving. As shown above you’re less likely to have an accident on a motorway in any event.

Without more detail I’m not sure their statistics tell us anything meaningful, or prove that we’re safer with Autopilot on.

 

[Wol747]

They’ve been transparent in their methodology. Ignoring the x10 reference from Elon they’re saying Autopilot it about twice as safe. It’s just not a like for like comparison when you’re comparing ‘mostly motorway’ Autopilot with ‘all roads’ human driving. As shown above you’re less likely to have an accident on a motorway in any event.

Without more detail I’m not sure their statistics tell us anything meaningful, or prove that we’re safer with Autopilot on.

I'm suspicious by nature and that quote doesn't explicitly say whether a crash following an intervention is counted as "autopilot" or "driver", just that it's counted as one of the crashes. Certainly the five second rule does make one thing clear - what's actually being counted - if not which side of the ledger!

 

[pgkevet]

We also do not differentiate based on the type of crash or fault, and in fact, more than 35% of all Autopilot crashes occur when the Tesla vehicle is rear-ended by another vehicle.

That gave me a chuckle. Phantom braking rulz OK

 

[NewbieT]

Another perspective (and some man maths you can shoot at). There are c40,000 Tesla cars in the UK now. Average mileage for a car is about 7,000 (or 18,000 for company cars).

To choose a number let’s say a Tesla averages 1,000 AP miles a year. Not everyone has/uses AP. 35k x 1k = 35 million AP miles a year in the UK.

If AP has one crash for every 4 million miles driven (as reported by Tesla) we can expect AP to be in control for say 8 accidents over the next year in the UK. I hope it’s not me as my AP miles def higher than the average above! Given the hype I expect we’d hear about the accidents too - 1 per month? Eek.

 

[Gatsojon]

They’ve been transparent in their methodology. Ignoring the x10 reference from Elon they’re saying Autopilot it about twice as safe. It’s just not a like for like comparison when you’re comparing ‘mostly motorway’ Autopilot with ‘all roads’ human driving. As shown above you’re less likely to have an accident on a motorway in any event.

Without more detail I’m not sure their statistics tell us anything meaningful, or prove that we’re safer with Autopilot on.

This is the issue with these numbers for me; we are simply not comparing like for like.

 

[pdk42]

They’ve been transparent in their methodology. Ignoring the x10 reference from Elon they’re saying Autopilot it about twice as safe. It’s just not a like for like comparison when you’re comparing ‘mostly motorway’ Autopilot with ‘all roads’ human driving. As shown above you’re less likely to have an accident on a motorway in any event.

Without more detail I’m not sure their statistics tell us anything meaningful, or prove that we’re safer with Autopilot on.

Yes, agreed. I’ve always thought this “AP is safer” statistic is potentially misleading.

 

[Mrklaw]

Another perspective (and some man maths you can shoot at). There are c40,000 Tesla cars in the UK now. Average mileage for a car is about 7,000 (or 18,000 for company cars).

To choose a number let’s say a Tesla averages 1,000 AP miles a year. Not everyone has/uses AP. 35k x 1k = 35 million AP miles a year in the UK.

If AP has one crash for every 4 million miles driven (as reported by Tesla) we can expect AP to be in control for say 8 accidents over the next year in the UK. I hope it’s not me as my AP miles def higher than the average above! Given the hype I expect we’d hear about the accidents too - 1 per month? Eek.

I guess the point being - if its safer (big, big if) - then that 'eek' is still going to be fewer accidents than those driving without. We can't expect FSD cars to never be in an accident, just that we would expect them to be in fewer (hopefully far fewer) than non self driving

 

[Tony Hoyle]

I disable AP when things get complex, like around here where several roads join the motorway at once and cars weaving all over is the norm. I don't even engage it outside motorways or divided A roads.. So of course it's going to be in fewer accidents because it never has to face difficult driving decisions.

It's probably true to say that the combination of AP + Human is safer than human on their own, but AP on its own would probably crash within minutes.

 

[GRiLLA]

Basically it's the US intersection that accounts for a great deal of the difference, generally

I disable AP when things get complex, like around here where several roads join the motorway at once and cars weaving all over is the norm. I don't even engage it outside motorways or divided A roads.. So of course it's going to be in fewer accidents because it never has to face difficult driving decisions.

It's probably true to say that the combination of AP + Human is safer than human on their own, but AP on its own would probably crash within minutes.

I think 'within minutes' is clearly exaggeration, when I'm using Autopilot I almost never have to interfere with it functioning, sometimes for hours on motorway journeys. I do agree that I would take over for anything remotely complex, exits mainly, and I rarely use it on A roads and never on anything else. Even still, it's of significant value to me, and even adding a small extra step like fully automated lane changing would make a huge difference. I certainly feel safer, I'm able to have much more awareness of what's happening around me without the hassle to steering and accelerating.
As drivers we tend to over-rate our personal skill and ability behind the wheel, as accidents clearly show. While yes I'm almost certainly a better driver than AP most of the time, sometime for brief moments I'm not, my attention is distracted or I make a poor decision. AP doesn't suffer from this.

 

[GeorgeSymonds]

It's total bunkum really.

What Tesla haven't said is compared to the same quarter last year, its got worse, in fact at best, its plateau'd gpt 18 months despite hundreds of thousands of cars out there helping to refine the models and multiple updates..

Tesla also publish stats for Teslas without passive safety, Tesla with passive safety and active AP. No passive safety is massively safer than the average which probably tells you a lot about the demographics of the driver and the type of usage.

Passive safety is about 30% higher than a Tesla without passive safety, you expect it to be better as it has various warnings etc. so its welcome but but not orders of magnitude better

Active AP is said to be twice as good as passive, but, and its a massive but and one others have alluded to, AP is only really used on motorways and dual carriage ways (which are 4x safer than the worst roads in the uk) and only really used in good conditions (both weather and traffic permitting). I don't know anyone that turns on AP when they think the going is getting too difficult for them. Invariably, tricky bits like junctions, roundabouts, turning off roads, traffic lights, you turn it off. I think there's a stronger case to say Active AP is less safe than just passive AP.

I also saw an article in a different forum that had the Nissan Leaf outperform its segement by 9-11x depending how you measure it for deaths,

https://www.iihs.org/api/datastoredocument/status-report/pdf/55/2

I think that probably says something about both the demographcs of the owners and that stats can get unreliable.

 

[GRiLLA]

I think all we are proving is that the US isn't really much like the UK. In a survey 2% of US drivers admitted they had once tried to run somebody else off the road due to driving disagreement. Anyone wonder why having the Cyber Truck bulletproof is a good sales tactic.