FSD Beta 10.69

[sleepydoc]

Seems like it must be able to, otherwise it would be getting into collisions all the time. I guess we’ll find out soon enough.

Doesn’t seem like an extraordinarily difficult problem. Just don’t hit other cars?

Yeah - car entering your lane => brake. In this case braking likely would have prevented the accident completely.

 

[Daniel in SD]

Seems like it must be able to, otherwise it would be getting into collisions all the time. I guess we’ll find out soon enough.

Doesn’t seem like an extraordinarily difficult problem. Just don’t hit other cars?

It does seem so but maybe Germans are much better at changing lanes?
That does look like a very easy collision to avoid, no prediction required.
I’m not convinced Mercedes is going to release their system here. They’ve only reported one collision to the CA DMV so they’re definitely not doing much testing.

 

[AlanSubie4Life]

In this case braking likely would have prevented the accident completely.

That does look like a very easy collision to avoid, no prediction required.

Maybe this is why 10.69.3 has not gone to wide release.

A few small regressions. (Yeah, I know this is not a regression.)

 

[Bladerskb]

Yep, absolutely no precedent for these kinds of accidents with autonomous systems: This Arizona college student has taken over 60 driverless Waymo rides

"Most sideswipe crashes (eight out of ten) involved the other vehicle changing lanes into the Waymo vehicle's lane. In one of the cases where the Waymo changed lanes, Waymo says the other car was traveling 30mph above the speed limit."

This wasn't a side-swipe, nor was this simulated. Stop spreading misinformation in order to absorb FSD Beta of any accountability.
Its pathetic. This was a regular lane change that happens thousands of times daily in traffic jams. If the driver wasn't tired (sleepy and not paying attention) this wouldn't have happened.

 

[AlanSubie4Life]

It’s mysterious to me how FSD Beta will stop for some things, but not others. Mysterious, but not surprising, to be clear.

So curious about why it doesn’t work in some trivial cases.

 

[flutas]

As far as I can tell the Model 3 is RWD until it senses slip. I actually put my car on a lift to verify a few years ago. I believe the A7 is true AWD (center differential, not some clutch nonsense).
Though last time I drove my 3 in the snow it was much improved. I think they may sense slippery conditions now and proactively engage front motor torque. The rear motor is a more efficient permanent magnet motor so that would be a reason to normally be RWD.

I believe (trying to find where I saw it again) that it's 70/30 back/front normally, but if it detects slip it goes to 50/50.

 

[JulienW]

I believe (trying to find where I saw it again) that it's 70/30 back/front normally, but if it detects slip it goes to 50/50.

Actually there is a permanent magnet motor in the rear and an induction motor in the front. For efficiency Tesla "free rotates" the induction motor (can't do this with a permanent magnet motor). The front is almost always in "free rotate" mode except when needed for acceleration or traction. So almost all the time the 3/Y is rear wheel drive only but it can never be a front wheel drive only.

EDIT Here is a video you can watch and see the front motor usage. Here is a screen shot under moderate acceleration and you can see the front has no power.

 

[swedge]

It’s mysterious to me how FSD Beta will stop for some things, but not others. Mysterious, but not surprising, to be clear.

So curious about why it doesn’t work in some trivial cases.

Alan, in the collision video, watch the distance to the car ahead. It looks like that distance starts our fairly constant, but increases rapidly, starting just before the collision. This would suggest that either the car ahead accelerated or Raj's car slowed down. Is it possible that Raj's car did actually brake and slow abruptly just before the collision?

Either way, it seems to me that any self driving car can still get hit when other drivers do dumb things like this. On the other hand, good defensive driving is clearly something FSD should be tutored on.

SW

 

[sleepydoc]

It does seem so but maybe Germans are much better at changing lanes?
That does look like a very easy collision to avoid, no prediction required.
I’m not convinced Mercedes is going to release their system here. They’ve only reported one collision to the CA DMV so they’re definitely not doing much testing.

I thought one of the restrictions on MB’s system was that it had to stay in the same lane?

 

[EVNow]

He followed up in later Tweets:

Also claimed he didn't have his foot on the accelerator, but he also said he was driving while tired, so anything's possible.

Yes, now I see those.

FSDb should have caught it. He should have been able to intervene as well - probably tired and distracted.

Ofcourse, in terms of driving rules/law, the other guy is probably at fault.

 

[sleepydoc]

Alan, in the collision video, watch the distance to the car ahead. It looks like that distance starts our fairly constant, but increases rapidly, starting just before the collision. This would suggest that either the car ahead accelerated or Raj's car slowed down. Is it possible that Raj's car did actually brake and slow abruptly just before the collision?

Either way, it seems to me that any self driving car can still get hit when other drivers do dumb things like this. On the other hand, good defensive driving is clearly something FSD should be tutored on.

SW

I thought it looked like the reverse- the car ahead accelerated a bit, temporarily creating an opening. A car tried to merge into the opening while Raj’s car simultaneously accelerated to close the gap.

Another factor against the other driver was that s/he didn’t signal properly. (Still doesn’t excuse FSD but is another strike against the other driver)

 

[Daniel in SD]

I thought one of the restrictions on MB’s system was that it had to stay in the same lane?

Yeah. In Gali’s collision he stayed in his lane and still hit another car that cut him off. An L3 system would need to avoid that collision unless customers are willing to have their car spend half the time in the body shop. haha

 

[AlanSubie4Life]

It looks like that distance starts our fairly constant, but increases rapidly, starting just before the collision. This would suggest that either the car ahead accelerated or Raj's car slowed down.

I think this description is a bit off, but I think I see what you mean; I see the nose of the car rise just before the collision as the car accelerated (or it was just a large bump in the road but I didn’t see that, no such bumping from lead cars).

So to me it looks like Raj (Gali?) pressed the accelerator either:
1) to catch up with lead traffic because FSD was plodding
2) he meant to hit the brake. Because FSD was not slowing down in response to merging traffic

The nose rise was too much to be explained by FSD. This is also observed by others in the Twitter comments.

Anyway without cabin footage, with likely accelerator application, we won’t know whether FSD saw the vehicle (there would have been furious beeping if it wanted to slow down). If it did not see it, there would be no warnings of course.

Just hard to say. Anyway just have to pay attention. And seems pretty clear FSD did not respond appropriately (in the fraction of a second before accelerator application, I did not see the nose dip).

 

[AlanSubie4Life]

while Raj’s car simultaneously accelerated to close the gap

Looked way too snappy for FSD (without encouragement).

the other driver was that s/he didn’t signal properly.

Yes for fault seems bad to only provide signal a fraction of a second before making your move.

Looks to me probably could have signaled a couple seconds earlier and it might have made no difference to the outcome. Just depends on attentiveness.

Upon video review, this vehicle was at fault for driving like dung; never would have happened otherwise:

 

[willow_hiller]

This wasn't a side-swipe, nor was this simulated. Stop spreading misinformation in order to absorb FSD Beta of any accountability.
Its pathetic. This was a regular lane change that happens thousands of times daily in traffic jams. If the driver wasn't tired (sleepy and not paying attention) this wouldn't have happened.

You didn't read the ArsTechnica article at all, did you? 18 actual collisions, and a further 27 simulated crashes that "would have occurred if the safety drivers hadn't intervened."

How could a safety driver possibly intervene in a sideswipe like you've illustrated? A safety driver wouldn't see a vehicle approaching it directly from the side, and certainly wouldn't have time to intervene.

This accident, just like with Waymo, was a sideswipe with enough time to intervene.

 

[AlanSubie4Life]

How could a safety driver possibly intervene in a sideswipe like you've illustrated? A safety driver wouldn't see a vehicle approaching it directly from the side, and certainly wouldn't have time to intervene.

Huh? Absolutely tons of time in the type of situation illustrated. Very confused here.

Happens all the time, nearly all collisions avoided.

In the situation illustrated, can even avoid it before anything (movement in lanes) happens! (In the illustration, the lane movement was likely evident at least 5 seconds in advance of it occurring.). Though that’s an extreme case. In general someone in an adjacent lane suddenly moving lane position without warning can often be avoided. Also don’t be there (key preventative measure).

 

[EVNow]

Doesn’t seem like an extraordinarily difficult problem. Just don’t hit other cars?

Not so trivial. FSD needs to figure out the probability of the other car trying to get into this lane even though it will result in an accident unless the ego car brakes. Most of the time the other cars won’t change the lane … but sometimes they do.

I think more importantly, once the other car definitely started to change the lane, ego car should slam on the brake (ofcourse could have been rear-ended for that).

Unfortunately no in cabin camera capture. I feel there is more to this than just FSD doing it’s stuff … at a minimum, FSD would have given up and asked for intervention…. With blaring red hands.

 

[momo3605]

Seems like it must be able to, otherwise it would be getting into collisions all the time. I guess we’ll find out soon enough.

Doesn’t seem like an extraordinarily difficult problem. Just don’t hit other cars?

I swear FSD Beta has SOME sense of cars with turn signals on, i've noticed my car slow/phantom brake if a car has their turn signal on and moving into your lane in front of you, but maybe it's just coincidence or something else.

 

[AlanSubie4Life]

Not so trivial. FSD needs to figure out the probability of the other car trying to get into this lane even though it will result in an accident unless the ego car brakes. Most of the time the other cars won’t change the lane … but sometimes they do.

In this case, since the car signaled immediately before moving, seems fairly trivial to respond as soon as the car makes the move. There was lots of time and an eternity of notification for the computer (about half a second plus or minus, basically forever).

once the other car definitely started to change the lane, ego car should slam on the brake (ofcourse could have been rear-ended for that).

No need, just a gradual slowdown was all that was needed; not a difficult case here.

No need for hysterics or unpleasant backside attacks.

I’m nearly positive that going to zero pedal input about 0.5 seconds after the vehicle makes its move (either FSD or human driver) would have meant no collision. No need for friction braking. Tough to tell for sure in these videos, but looks quite possible.

I feel there is more to this than just FSD doing it’s stuff

Yes it looks like there was driver input, but after the point when FSD should have reacted, so does not absolve FSD.

 

[JulienW]

I swear FSD Beta has SOME sense of cars with turn signals on, i've noticed my car slow/phantom brake if a car has their turn signal on and moving into your lane in front of you, but maybe it's just coincidence or something else.

No and this will be somewhat controversial anyway. Many people turn them on and then ride with them on. Then of course BMW's don't have signal lights. So Beta would have hard time determining if the signal was correct, left on or not used at all.