FSD Beta 10.69

[AlanSubie4Life]

I have it was probably going faster and it was just this Monday.

View attachment 866318

No doubt I'm a little overly passionate but as a LONG time runner I have been in the mix a LOT and humus can be indecisive, inconsiderate, distracted and/or malicious of pedestrians. At least Beta has no passion even if not perfect.

I see your problem: you shouldn’t have been in the middle of the intersection and the bus lane. Sidewalks and crosswalks are safer.

I’ve run a lot over the years too though I avoid city streets when possible. Right-turning traffic is hazardous, but left-turning too due to a-pillar blind spots at “constant bearing, reducing range.”

 

[sleepydoc]

Apparently you have never been hit by a car going 3mph.

15 MPH is an incredibly appropriate rate of speed for a roundabout. I’m quite sure if the car was going 3 MPH through the roundabout you’d be complaining that it was going at an “extremely slow rate of speed”

 

[JulienW]

I see your problem: you shouldn’t have been in the middle of the intersection and the bus lane. Sidewalks and crosswalks are safer.

I’ve run a lot over the years too though I avoid city streets when possible. Right-turning traffic is hazardous, but left-turning too due to a-pillar blind spots at “constant bearing, reducing range.”

GPS track is off or not that accurate. Also I was in the crosswalk but DID end up well into the bus or even traffic lane while hood riding.

Also that brings up another constant Beta flub. Not recognizing and often driving in those same Bus lanes.

 

[AlanSubie4Life]

GPS track is off or not that accurate

Yes of course (as I clearly indicated!).

Terrible

Terrible

I think our disagreements (there is none here actually AFAICT) center around the use of this word. For me in this context it means “suboptimal” (perhaps that language you can agree with?). For you “terrible” means “failure.” I agree both of your cases are “terrible” (failures).

The video above shows a suboptimal stop when yielding to a pedestrian (“terribly executed” in my language).

 

[Dewg]

Also that brings up another constant Beta flub. Not recognizing and often driving in choses same Bus lanes.

This needs to get resolved, and can be resolved by either map data updates, or sign reading / lane marking reading for the planner. I wonder if generic map data (like TomTom, and OpenMaps) has information on bus-only lanes, like Lane #3 is flagged Bus-Only?

 

[Daniel in SD]

15 MPH is an incredibly appropriate rate of speed for a roundabout. I’m quite sure if the car was going 3 MPH through the roundabout you’d be complaining that it was going at an “extremely slow rate of speed”

Sure, but why accelerate only to slam on the brakes a second later?
The pedestrian is obviously spooked by it and I’m sure it wasn’t all that comfortable for the driver.

 

[AlanSubie4Life]

I’m quite sure if the car was going 3 MPH through the roundabout you’d be complaining that it was going at an “extremely slow rate of speed”

Why on earth would I complain about that, since there is a pedestrian clearly crossing and anyone entering the roundabout can see them, so would appropriately react to other vehicles’ reactions to said pedestrian?

Obviously I would complain about that behavior if it were slow off the mark for no reason.

Makes perfect sense. Just drive normally.

Obviously it has no idea how to handle roundabouts, so I am not surprised by any of the terrible behavior but still seems like they could stop smoothly.

 

[Dewg]

I went back and watched it frame by frame (there are 30 frames per second) - it's not as bad as I thought. From a time standpoint, the ped was detected at 1:46:30, and the decision was made to slow/brake for them at 1:47:40. That's 1.03 seconds to detect the ped, determine their intent in the crosswalk and begin slowing.

Where I think people are being critical is that the car didn't detect the ped until 1:46:30, when we humans could see the ped several seconds earlier. But given the 1 second, I'm a bit more impressed.

 

[AlanSubie4Life]

when we humans could see the ped several seconds earlier.

As I mentioned originally, so could FSD; it is extremely capable. Don’t rely on the visualizations, by the way - remember they lag by 0.5 seconds or so (EDIT: looks like about 0.3 seconds right now) and they may not reflect what the car can see (impossible for us to know right now). We can use them as a useful reference for timings as you did but they are not definitive as far as calculating reaction time (they provide minimum value + display latency after detection of a displayable object).

That's 1.03 seconds to detect the ped, determine their intent in the crosswalk and begin slowing.

This is actually quite poor! And remember to add about 0.5 (EDIT: 0.3) seconds!

One extra second vs. a poor-quality human (who failed to see the peds before entering the roundabout) makes a huge difference and is a poor performance.

I have to conclude that many people here have very low opinions of their own driving (unclear whether justified or not; I suspect often not, though my wife frequently complains about slow Tesla drivers so maybe you guys are onto something).

 

[AlanSubie4Life]

didn't detect the ped until 1:46:30

Ackchyually, it was at 1:46 and 4/30ths until initial display of white outline, or about 1:46.1. So detection presumably occurred at 1:45.8 or so. It was still going just 8mph (just 30% of the peak kinetic energy displayed in this event).

and the decision was made to slow/brake for them at 1:47:40.

How do you figure (I assume it is because the pedestrian was highlighted in dark at that time)? It was applying power at 1:48:00 (and then started slowing about two frames later). I think it's reasonable to assume that decision time to actual removal of power is so short it is irrelevant to the timing here:

So now that I've checked it myself, it's way way worse than even the quite poor 1.03 second of reaction time you claimed; over 2 seconds from detection to slowing. This is of course typical of what we see from FSD Beta in all circumstances (haven't yet seen a superhuman reaction (meaning less than 0.3 seconds, though sprinters can apparently react in 0.1 seconds to auditory stimulus) demonstrated, though I expect it will be coming before the end of the year).

And again, this is all assuming quite poor performance where FSD Beta somehow failed to detect pedestrians on the opposite side of the roundabout well in advance of visualization (it wasn't necessary for it to do that to avoid the extreme level of dysfunction demonstrated here, though).

Now, we have not seen an accident either, so it is possible large but safe delays are intentionally produced in order to provide jerk and unpleasantness, as a means of making the safety drivers less confident in system abilities (and giving them an opportunity to disengage), and to make VRUs more alert to the dangers of robotaxis. Though I doubt it, given the increased risk of a rear-end collision amongst other things.

 

[Dewg]

How do you figure (I assume it is because the pedestrian was highlighted in dark at that time)?

Watch the acceleration line, it's increasing as the car enters the roundabout, then stays steady until 1:47:40, where it begins to drop, indicating it's beginning its braking.

 

[AlanSubie4Life]

Watch the acceleration line, it's increasing as the car enters the roundabout, then stays steady until 1:47:40, where it begins to drop, indicating it's beginning its braking.

No, that's not how this works. Beginning to drop on this bar does not mean the car is decelerating!!! That only happens when forces are matched. It's natural to have a large line, followed by a shorter line, with much slower subsequent acceleration (the car is still accelerating!), especially on starts like this. Maintaining speed on level ground at 14mph requires very small amounts of power (less than about 2kW, 143Wh/mi, not including non-motive power), even with wet roads.

In fact, the speed of the vehicle INCREASES at 1:47 and 19/30ths (1:47.63), and likely is still increasing after that point, since it is unlikely that the 13mph=>14mph change captured the exact frame where forces matched.

It is entirely possible that it identified the VRU and decided to stop accelerating at around that point rather than powering through the apex. So that would be a "reaction." So we could charitably say that it stopped accelerating at 1:47.7 or so. But that's not the "decision to slow."

 

[EVNow]

So now that I've checked it myself, it's way way worse than even the quite poor 1.03 second of reaction time you claimed; over 2 seconds from detection to slowing.

Humans would not enter the roundabout - just let the peds cross before entering the roundabout.

 

[Daniel in SD]

I have to conclude that many people here have very low opinions of their own driving (unclear whether justified or not; I suspect often not, though my wife frequently complains about slow Tesla drivers so maybe you guys are onto something).

It makes sense that the earliest adopters of automated driving technologies would be the worst drivers.

Humans would not enter the roundabout - just let the peds cross before entering the roundabout.

But then you’d be blocking people who want to go straight or turn left.

 

[EVNow]

But then you’d be blocking people who want to go straight or turn left.

Stopping inside the roundabout would be worse. Besides I'm not "blocking" - I'm waiting.

 

[Daniel in SD]

Stopping inside the roundabout would be worse. Besides I'm not "blocking" - I'm waiting.

Yeah you don’t stop in the roundabout, you stop where FSD stopped. You just do it much more slowly to not jerk your passengers around and not spook the pedestrian.

 

[EVNow]

Yeah you don’t stop in the roundabout, you stop where FSD stopped. You just do it much more slowly to not jerk your passengers around and not spook the pedestrian.

Looks to me the car was partly blocking the roundabout - at least it's very close.

 

[AlanSubie4Life]

Looks to me the car was partly blocking the roundabout - at least it's very close.

No, it's not, and it's really not even close. The car behind is blocking the roundabout though.

 

[EVNow]

No, it's not, and it's really not even close. The car behind is blocking the roundabout though.

I'll watch again later and post. On the phone now...

 

[PACEMD]

For you “terrible” means “failure.”

I think in this case "terrible" is defined as the nature of this conversation.