Tesla Vision earns IIHS "Superior" rating avoiding crash in *all* pedestrian tests

2022 Model Y received TOP SAFETY PICK+ today with both front crash prevention tests (vehicles and pedestrians) getting Superior rating. Tesla has consistently gotten Superior rating for vehicle-to-vehicle, but even with a Tesla Vision re-test with updated software back in June, the vehicle-to-pedestrian score "only" scored Advanced.

However with the latest Tesla Vision tested in the Model Y and re-tested with Model 3, not only did IIHS rate it Superior, all pedestrian tests avoided collisions including traveling at 25mph with a small dummy running out from behind parked vehicles obstructing view. Previously with radar Autopilot, the crossing child test only reduced the speed by 5mph.

What other safety systems can completely avoid collisions in all IIHS tests? And of those, how many vehicles have it as a standard feature? I wonder how long until even Tesla vehicles with radar get switched over to Tesla Vision to improve safety.

 

[linux-works]

except your whole premise is incorrect.

removing degrees of freedom does not help.

but hey, drink the koolaid. its yummy.

 

[AndreP]

Teslas are not cheap, so I don't know if it's fair to hold these features up as standard unless comparing specifically to comparably-priced models/trims. But I think the main complaint right now for 2021 models with Vision-only would be extraneous braking, not the absence of braking.

 

[JulienW]

.....Of course the issue with these tests is that you know the questions before the test! I always wonder how well these systems work in the real world.

That's not really BIG problem. The big problem is as a pedestrian you don't get to choose one of these car's driver's as the one to not be paying attention to you.

 

[EVNow]

Teslas are not cheap, so I don't know if it's fair to hold these features up as standard unless comparing specifically to comparably-priced models/trims.

Nope.

If "saving lifes" is actually as important as people claim it is - this should be the minimum standard.

ps : All auto majors have "voluntarily" agreed to include collision avoidance far in the future. After the current executives have made their millions, bought their yachts and retired.

https://www.cars.com/articles/automakers-safety-officials-make-crash-avoidance-systems-standard-by-2022-1420684105016/

Don't know whether the 2022 is still the case or has been moved.

 

[Terminator857]

... removing degrees of freedom does not help. ...

Can you expand on this? In reference to vehicle to pedestrian tests: Do you mean: Neural networks excel in fixed tasks? Car companies can easily game these tests by specifically programming for them. Did Tesla add the video from prior year test to the neural network training data with labels? If the tests were not "fixed" and had some randomness to them, then I would have more faith in the tests. Hopefully night time tests are coming. Rain, fog, sun in the eyes/cameras also? How well does it detect a person wearing black clothes on a black street at dusk?

Trivia:
Model 3 roof: Strength-to-weight ratio 5.85
Model Y roof: Strength-to-weight ratio 4.42

 

[Mardak]

except your whole premise is incorrect

Which are you referring to? That Tesla might switch radar vehicles to Tesla Vision?

I'm well aware of issues with Tesla Vision from using FSD Beta on highways such as unnecessary "faster lane" changes because it incorrectly detected a slow obstacle ahead. But if Tesla is looking to improve its safety numbers and Tesla Vision can do that even with the driver/passenger discomfort, they could choose to save more lives for that tradeoff.

Of course, Tesla can make more improvements to Tesla Vision to increase safety even more and/or reduce its problems, but there's still a question of when the switch to no-radar should happen.

 

[EVNow]

Its funny the concern trolls concentrate on this company.

https://twitter.com/x/status/1473652299499253763

 

[S4WRXTTCS]

Curious to see how it will do with pedestrian tests done at night.

That's where a lot of AEB systems fail.

I believe the Euro NCAP does, and AAA has it as well.

The Euro NCAP one was just tested for the 2021 Model Y, but I don't see the actual report available yet.

Here is are the results from a fairly difficult AAA test where nothing did really well, but that was 2019.

AAA Warns Pedestrian Detection Systems Don’t Work When Needed Most

Study finds safety systems fail at night when the majority of pedestrian vehicle fatalities occur

newsroom.aaa.com

 

[JulienW]

 

[JHCCAZ]

One interesting point is that the IIHS safety ratings are associated with the model year, which made perfect sense until now. With FW updates able to improve all cars on the HW3 platform, seems it's time for IIHS and other bodies to adapt in publishing revised results based on the HW+SW complement that will be more and more independent of the Model and Year association. This is mostly a Tesla thing now, but other makes will follow.

It does matter that cars can improve after they're first sold.

...all pedestrian tests avoided collisions including traveling at 25mph with a small dummy running out from behind parked vehicles...

I feel lucky I made it this far, surviving my days as a small dummy running around when cars had no dummy avoidance technology...

 

[username123]

What other safety systems can completely avoid collisions in all IIHS tests? And of those, how many vehicles have it as a standard feature?

Mazda 3 for half the price has had it as standard since 2019.

2022 Mazda 3 4-door hatchback

2022 Mazda 3 4-door hatchback - small car

www.iihs.org

Honda Civic can avoid the crashes too

 

[BigNick]

Considering the radar-removed Tesla will brake for shadows and other things that aren't even present, it should be able to avoid dummies and other solid objects!

Tesla needs to fix the phantom braking issue, full stop.

 

[JulienW]

Considering the radar-removed Tesla will brake for shadows and other things that aren't even present, it should be able to avoid dummies and other solid objects!

Tesla needs to fix the phantom braking issue, full stop.

No one would disagree that phantom needs to be reduced/fixed and will over time. However it is somewhat of a contradiction since if Tesla eliminated all false positives today it would increases false negatives. While false positives are inconvenient false negatives can result in injuries or deaths and of course failure of the IIHS test.

 

[SageBrush]

Mazda 3 for half the price has had it as standard since 2019.

If by 'it' you mean a superior rating, then yes.
If you mean Tesla level, then no. At 37 mph the Mazda issued a warning 1.8 seconds before a collision would have occurred. Tesla vision warned 2.7 seconds before collision.

This difference is huge -- just imagine streets with poor traction.

 

[BigNick]

No one would disagree that phantom needs to be reduced/fixed and will over time. However it is somewhat of a contradiction since if Tesla eliminated all false positives today it would increases false negatives. While false positives are inconvenient false negatives can result in injuries or deaths and of course failure of the IIHS test.

False positives leading to panic-level braking, as though an object were in the road, can also cause injuries and deaths.

 

[sleepydoc]

I remember reading a post/story recently about how poorly Tesla (and most/all other makes, to be fair) performed on pedestrian avoidance maneuvers. I wonder what the difference was in this test? Is it a never software version? Was the test itself different?

The other question/point to make is with any detection system there are tradeoffs between sensitivity and specificity. The more accurate a system is, the smaller the tradeoff, but in any system increasing the sensitivity will also increase the number of false alarms. Measuring the number of false alarms is difficult, because they are relatively infrequent and unpredictable but false AEB events are a known issue for Tesla. (again, to be fair, they occur in other makes as well, they just don’t seem to be quite as prevalent.)

The million dollar question is how do you balance safety with an effective AEB system vs the danger and inconvenience of false activations and what is the ‘right’ balance? (The $10M question is how to make a system that does both.)

 

[Mardak]

Mazda 3… Honda Civic can avoid the crashes too

Thanks. I wonder if both have relatively high phantom braking too as others pointed out achieving high true positive often comes with higher false positives. Similarly, some people say they prefer no false positives, but that likely comes with a human cost in terms of safety. Hopefully Tesla can significantly improve precision while maintaining or even improving recall with FSD Beta stack then deploy it to the whole fleet even with standard Autopilot.

 

[JulienW]

False positives leading to panic-level braking, as though an object were in the road, can also cause injuries and deaths.

But can you site any evidence that this is happening beyond anecdotal? We know that about 20 VRU are killed each day in the US. Don't know of any statistics on rear end fatalities caused by false positive/Phantom braking.

 

[Huskyf]

Mazda 3 for half the price has had it as standard since 2019.

2022 Mazda 3 4-door hatchback

2022 Mazda 3 4-door hatchback - small car

www.iihs.org

View attachment 773462

Honda Civic can avoid the crashes too

Yes but for Tesla Vision in Parallel adult test they provide a warning 2.8 before impact