Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Autopilot safety question

This site may earn commission on affiliate links.
I saw someone say in a reply on another thread that the Autopilot Safety Feature would not bring a car to a full stop in an emergency. It will only slow the car to reduce damage. Is this true? If so Tesla is behind other manufacturers in the area. I’ve seen commercials from other companies on television demonstrating cars stopping to avoid hitting objects.

I looked quickly on the Tesla site and did not see where is said one way or the other. Thanks for any clarifications.
 
Thanks. I was thinking it would have been a rather large miss by Tesla if it would not bring the car to a full stop.

Bringing the car to a full stop is actually not all that important IMO, with the sole exception of the driver becoming incapacitated / deceased, which is an edge case.

If the car initiates braking, you will probably snap your attention back to what's going on, and either agree with the car and slam on the brakes, or disagree with the car and hit the accelerator. Since these systems do have false alarms given today's technology, it's less hazardous to suddenly slow by 25mph vs coming to a complete stop at highway speeds.

Overall though, collision avoidance is an assistance system. Even if it triggers and you agree with it, the expectation is you will take over braking, and not just sit there idly assuming that the car will stop on its own!
 
I'm pretty sure the lawyers prevent any manufacturer from claiming that the car will always come to a full stop. It's much safer to weasel out a bit and say thinks like "reduce speed" and "lessen the severity of an impact" than to say "we'll stop your car for you and prevent a collision."
 
  • Like
Reactions: Tiberius
I'm pretty sure the lawyers prevent any manufacturer from claiming that the car will always come to a full stop. It's much safer to weasel out a bit and say thinks like "reduce speed" and "lessen the severity of an impact" than to say "we'll stop your car for you and prevent a collision."
The Tesla manual doesn't guarantee to bring the car to a full stop and completely avoid an accident (just "minimize damage") so the lawyers have had their say. However, there are well documented dashcam videos and many anecdotal accounts of full stop and no damage.
 
I'm pretty sure the lawyers prevent any manufacturer from claiming that the car will always come to a full stop. It's much safer to weasel out a bit and say thinks like "reduce speed" and "lessen the severity of an impact" than to say "we'll stop your car for you and prevent a collision."
I agree. This sounds like classic CYA on Tesla's part, and for good reason.
 
EAB was activated in the van video and did slow the car.
The manual reads that the EAB will slow the car to reduce serious risk of injury. The driver that hit the van was not injured, so it did what it was supposed to. It also beeped and alerted the driver, he just wasn't paying attention enough to take over and fully stop the car.
 
Then there's the time that AEB didn't do a single thing. So there's a very good reason why Tesla has a CYA clause. This feature is nice when it works, but really not guaranteed to work - conditions must be favorable.

The difference would be that the van was stationary. Generally single camera in model s isn't the best at sensing stationary objects. People over in the model s forum have discussed this at length.
 
The difference would be that the van was stationary. Generally single camera in model s isn't the best at sensing stationary objects. People over in the model s forum have discussed this at length.

Linked in the Youtube comments on that video by the Tesla owner in the video are two additional videos of him doing two "tests" of similar scenarios where the vehicle actually braked.

The system screwed up and appears to have screwed up in several of these recent crashes (even if they've been primarily driver-error crashes).
 
  • Informative
Reactions: Colsla
Linked in the Youtube comments on that video by the Tesla owner in the video are two additional videos of him doing two "tests" of similar scenarios where the vehicle actually braked.

The system screwed up and appears to have screwed up in several of these recent crashes (even if they've been primarily driver-error crashes).
Both of his similar tests are not high speed tests against a stationary object in the same lane after the vehicle in front exits the same lane. His second video was close but he's head on instead of offset, going too slow, and there are no obvious lane markings.
 
The two problems with that collision are:

(1) the van was stationary. Doppler radar cannot detect stationary objects, so only the camera was in play.

(2) the van was off center in the lane AND weirdly shaped/colored. Is the computer vision algorithm in the camera going to recognize that thing as a car, and determine its in your path? It turned out it eventually did, but too late for the driver to react.
 
The two problems with that collision are:

(1) the van was stationary. Doppler radar cannot detect stationary objects, so only the camera was in play.

(2) the van was off center in the lane AND weirdly shaped/colored. Is the computer vision algorithm in the camera going to recognize that thing as a car, and determine its in your path? It turned out it eventually did, but too late for the driver to react.

If the van is stationary and the car is moving toward it then radar and the doppler effect will work as expected. You also don't need movement for radar range detection. However you'd be detecting the leading vehicle until it suddenly changed lanes.
 
If the van is stationary and the car is moving toward it then radar and the doppler effect will work as expected. You also don't need movement for radar range detection. However you'd be detecting the leading vehicle until it suddenly changed lanes.

Except the patch antenna is single field of view, so a speed bump, hill, road sign, or guard rail would all be negative vehicle speed doppler returns (e.g. stationary objects). It is quite difficult to differentiate all the doppler returns going 0mph relative to your vehicle's motion, and often times that is handled in more sophisticated systems by lower range directional radar with limited field-of-view (or 3D contour mapping LIDAR) correlated with a camera.

Tesla's camera seems to get it right more often than not, which is a ton more than I can say about my Audi which got it right 0 times. But still, it's a pretty hard problem to solve.