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

Another tragic fatality with a semi in Florida. This time a Model 3

This site may earn commission on affiliate links.
Status
Not open for further replies.
Two reasons (at least.)

1. Often, the worse accident is caused by cars following those which have an accident! Suddenly stopping to zero is likely to cause additional accidents, often worse ones. (If EAP was driving, it would have probably already been braking ahead of time, so other drivers would be more aware and the additional panic stop is less likely to cause worse accidents.)

2. If EAP is not engaged and you're driving, then you have your foot on the pedals. During the time that the car brakes from 70 to 40 (or whatever) you've certainly moved your foot to the brake and can continue to brake IF THAT'S THE CORRECT ACTION. As the driver, you can choose whether to deflect left or right, or to just continue to to brake. Perhaps the obstacle in front of you has already crossed the road (deer? Cross traffic?) and you can see the road is now clear. If so, you do NOT want to continue to panic brake and cause someone behind to rear-end you or to panic-turn into oncoming traffic and cause an unnecessary fatality.

This is probably the best explanation so far.

Phantom breaking is an imperfection in obstacle detection and imaging processing. It does not explain why full stop is attempted when TACC is engaged but only 30 mph deceleration is attempted when TACC is not engaged.
 
  • Like
Reactions: OPRCE
Keenly aware of the original 2016 Joshua Brown accident, I keep tabs on my Model 3's reaction to similar situations. I work in a port city with lots of warehouse truck traffic. Take an example of a 2 lane undivided highway with a 35 mph speed limit. Don't get on my case about recommended usage... I just do it to evaluate AP's response to the conditions. Say a vehicle travelling in the opposite direction turns left across your path.
  • If it's a car, AP will slow down. Perhaps even to the point of over-reacting (slows too much, too early, and doesn't resume for too long after path is clear).
  • If it's the front half of a semi, AP will slow down.
  • If it's the rear (trailer) half of a semi, AP doesn't slow down. I've experienced this several times.
I think Tesla has put a lot of effort into making sure the driver is attentive. But if we're honest, the car still probably doesn't properly or sufficiently react to a semi trailer crossing its path.

Has anyone (safely) observed an appropriate AP response to this scenario?
 
On the plus side, this is confirmation AP2.5 has feature parity with AP1.

I personally can't wrap my head around why this accident happened. There is a user called verygreen who has posted videos of what the AP2.5 system detects, and I've never seen it fail to detect crossing traffic whether it was a car or a semi with trailer.

I believe a lot of that stuff is essentially alpha / beta / experimental detection, and not yet being used by AP for decision making - likely getting a lot of false positives and false negatives. Cross traffic isn't something officially handled by EAP right now. The latest update shows the computer is detecting stop lines, but we know EAP isn't reacting to them.
 
Phantom breaking is an imperfection in obstacle detection and imaging processing. It does not explain why full stop is attempted when TACC is engaged but only 30 mph deceleration is attempted when TACC is not engaged.

Because "not TACC" is, by design, less automation intrusion into the vehicle control. It is a back-up, rather than the primary control. Thus inherently it has less time/distance to brake, as it gives the driver more leeway. The default assumption is the driver knows what they are doing AND it also can't actually see far enough ahead at 50 mph or over to reasonably act at all, as turning up the sensitivity on the automation would create a lot of false positives out of the signal noise, thus braking for erroneous reasons (and potentially leading to read-ending accidents behind the vehicle).

TACC by explicit design takes on more responsibility to bring the vehicle to a stop. Its job isn't to avoid collisions, its job is to reduce energy of collisions. So it can, and really must, begin braking far sooner. Braking sooner means being far more likely to brake to a full stop ahead of a detected obstacle.
 
Whether it is official or not, my EAP will brake for crossing vehicles such as those turning left in front of me.

Does it do that at highway speeds? I suspect you were within Radar range to easily detect movement.

Previously, objects at a stand still coming into Radar were not flagged by AP1, hence hitting stopped fire trucks, and possibly the 2016 semi. I assumed this was still true, but people pointed to this being resolved with AP2+ with its elevation detecting radar. So I figured I was wrong, and it's been fixed. Now, I'm not so sure. Wasn't AP2 suppose to fix this exact issue?
 
  • Like
Reactions: OPRCE
On the plus side, this is confirmation AP2.5 has feature parity with AP1.



I believe a lot of that stuff is essentially alpha / beta / experimental detection, and not yet being used by AP for decision making - likely getting a lot of false positives and false negatives. Cross traffic isn't something officially handled by EAP right now. The latest update shows the computer is detecting stop lines, but we know EAP isn't reacting to them.

Yea, watching the videos that greentheonly has posted, it can be seen that the car can detect cross traffic as well, but reliability is the question.

@EngiNerd211 it would be interesting to see this tested on different types of long vehicles as well. thinking of buses, long box truck, open flatbed, etc. of varying lengths and ride heights. also just wondering did you ever observe where it did pick up the 2nd half the truck (like if it was lower, had coverings, etc)
 
  • Like
Reactions: OPRCE
This was a stainless (grey-ish) trailer crossing and blocking all southbound lanes to turn north on a divided but not limited access highway. The road surface was grey-ish and extended south as far as the eye can see. Sun position was not a factor.

That trailer was practically invisible visually. Worst-case camera scenario. Doesn’t mean the M3 did not see or react to the obstacle, but without TACC vehicle would only have slowed, not stopped, by design. We don’t know the actual details yet. But we know enough not to rely on the technology to keep us safe without a high level of human oversight.
 
  • Informative
Reactions: OPRCE
Does it do that at highway speeds? I suspect you were within Radar range to easily detect movement.
It happened just yesterday in a 50mph zone, nobody in front of me so rate of travel would have been 55mph. Road is 2 lanes each way and a center turning lane. Vehicle crossing from other side to my side, fully tangental to the road, in front triggered EAP to reduce speed.

However the weird part is that normally by the time the system has sorted out that there is an object crossing in front the vehicle is almost past, so a lot of the speed reduction happens when the vehicle has already passed the line of the right shoulder.
 
@EngiNerd211 it would be interesting to see this tested on different types of long vehicles as well. thinking of buses, long box truck, open flatbed, etc. of varying lengths and ride heights. also just wondering did you ever observe where it did pick up the 2nd half the truck
Agree. I don't happen upon conflict with long vehicles very often, only semi's in my case, and I tend to abort as soon as my Tesla accelerates toward the crossing vehicle. I agree that @verygreen's annotated video would be a good way to see if AP is properly identifying these situations.
 
  • Like
Reactions: OPRCE
  • If it's the rear (trailer) half of a semi, AP doesn't slow down. I've experienced this several times.
This actually makes sense to me. Since the Tesla radar is a Doppler radar there has to be movement for it to "see" something. If the front of the truck has already passed by the time the radar is in range then back half of the truck appears to non-moving object because it's moving across/perpendicular to the radar and there isn't moving target for it to register it as a vehicle. To the Tesla's radar it would be invisible because it's stopped from the radars perspective. I think EAP relies a lot on movement to tell vehicles from buildings, barriers, and other objects that by the road, but don't need to be concerning. Imagine a big square metal building that was ahead of you with the road curving around it. If all it saw was a radar echo it would have no idea whether the building is truck that is close or building that is a mile off. Starts to get complicated if you can't take movement into account.
 
Last edited:
  • Like
Reactions: OPRCE
I also wonder if the 'Auto' in Autopilot was meant to suggest automobile, automatic, or be ambiguous.

They used the technical definition of autopilot, which exists in avionics. Autopilot - Wikipedia

In an airplane, autopilot maintains speed and heading.

A car doesn't have a "heading," because it must stay in marked lanes, but it has a lane... So in a car, the equivalent is to maintain speed and lane.
 
They used the technical definition of autopilot, which exists in avionics. Autopilot - Wikipedia

In an airplane, autopilot maintains speed and heading.

A car doesn't have a "heading," because it must stay in marked lanes, but it has a lane... So in a car, the equivalent is to maintain speed and lane.
Has there ever been a midair collision at cruising altitude though? That's where the analogy sort of breaks down, avionics autopilot is way safer. I like the idea of changing the name to "copilot" as others have suggested.
 
Has there ever been a midair collision at cruising altitude though? That's where the analogy sort of breaks down, avionics autopilot is way safer. I like the idea of changing the name to "copilot" as others have suggested.
"Copilot" is taken by another automaker. Ford uses "Copilot 360" branding for one of their systems. Nissan uses ProPilot (not sure about the capitalization?).
 
Status
Not open for further replies.