Smart Summon Drove My Model Y Into A Tree - July 2021

[AlanSubie4Life]

The funny part is that it DOES build a map etc

It’s not clear to me how persistent the map is. I haven’t seen anybody analyze the memory of the system. I don’t think this area was well covered by ultrasonics (I also do not know if they use ultrasonics but it does not matter).

In this case the ultrasonics in the rear bumper (or cameras or whatever; it does not matter) should have seen the vehicle, memorized the location, and ensured that the area without ultrasonic coverage did not collide with the vehicle. Failing that (say the ultrasonics could not see the adjacent car when parked), the ultrasonics should have created a map when the car was initially parked, hours before summon was invoked. Just like a human. And do sanity checking to make sure the situation didn’t change while parked of course. Just like a human.

Anyway it does not matter. Obviously to anyone who has tried to use it, the system does not do most of this.

It’s clear to me with the sensor suite on the vehicle that the various collisions here could be avoided by really using the existing sensor suite plus memory. But Tesla isn’t bothering to do it (not sure why, probably limited resources to overcome the technical challenges - it’s one thing to have a sensor, it’s quite another to build an accurate picture of the environment (detect and identify narrow trees, poles, etc., things that are easy for humans) with vision and ultrasonics).

 

[Yoliber]

Ultrasonics have difficulty detecting narrow objects, the tree is pretty skinny. That's why a lot of these crashes are poles, trees, pillars, cones. I think 360 cameras would help with these low speed maneuvers. Looking at Tesla's vectorize birds eye view, I could see it having difficulty creating a map without holes within the first couple feet.

 

[JHCCAZ]

Quotes from the Y manual on Smart Summon:

WARNING: When you release the button to stop
Model Y, a slight delay occurs before the vehicle
stops. Therefore, it is critical that you pay close
attention to the vehicle’s driving path at all times
and proactively anticipate obstacles that the
vehicle may be unable to detect.

Depending on the quality of the
connectivity between the phone and Model Y,
there may be a slight delay between when you
release the button and when the car stops.

How long can the delay to stop be? Enough time to hit something I'll bet.

How are you supposed to know this: "proactively anticipate obstacles that the
vehicle may be unable to detect." Trial and error?

It seems to me that it should not be programmed a start or stop action toggle from the phone, and I didn't think it was because you're supposed to hold the app button down continuously.

It seems obvious that the feature should be polling the phone at least several times per second (ie an essentially imperceptible delay, not a "slight" delay), and if it doesn't receive positive confirmation that the phone is in active contact with the car, it just stops.

It's totally unacceptable that the "Finger off button, stop moving" condition requires active communication contact to react, as to a Stop "request". Clearly, Stop should be the default state unless actively confirmed otherwise, with maybe a 50ms poll rate and a 150ms time-out for simple noise filtering. If it is not (and the manual warning suggests it is not), then the way the app is programmed is an illusion. The app and UI make us think it's a hold-to-enable Deadman Control, but in the car software it's treated as Toggle On then wait for Toggle Off? Perhaps with a backup "safety" time-out that is dangerously long enough to be unsafe? Even as I write the question it seems kind of crazy that it would be implemented this way, if so it needs to be fixed right now - or the manual warning is wrong and needs to be rewritten.

This point is aside from the general discussion about whether Smart Summon should even be publically released before it is reasonably reliable. I think it will become so, but I'm not going to release it to myself until it is.

 

[DrDabbles]

On this situation they use radar and ultrasonic but not vision I think when Tesla vision is merge Summon become Excellent....

Your claim is that smart summon, which was famous for being the first option that used primarily vision for navigating, is not using vision? How do you suppose it (poorly) navigates around parking lots without vision?

 

[linux-works]

using wifi for control - real-time - of a moving CAR - its just insane.

I do wifi testing (as one of my roles) and I can attest that wifi is just NOT mission critical rated. there is way too much jitter and packet loss, not to mention security issues.

I completely disagree with using a phone to RC control a quadco^Hcar.

hell, I would not use a phone to pilot my $100 DIY drone. those use 2.4ghz but its not IP and not at all like WIFI other than the 2.4ghz bit.

I suppose any loss of communication to the phone means -immediate stop- but even that may not be what the user wants. an immediate stop, when he's not planning on it, can be just as bad!

didn't ANYONE at tesla think this thru? are there any adults left there? HELLO????

 

[linux-works]

Five years after that tweet and it's crashing into stationary objects within seconds of activation. We'll still be talking about its shortcomings in 2026.

in the year 2025
if tesla is still alive
if saplings can survive
we may find...

(again, GOML)

 

[emmz0r]

Isn't there a reasonable limit where tesla no longer can get away with putting beta on it?

When it does something unpredictable under perfect conditions means that you can never use it. So by that logic, is the feature truly released, if it never can be used in practice?

 

[qdeathstar]

Hey, in Tesla’s defense it was a small tree

 

[sperkin]

I only tried Smart Summon a couple of times while I was in the car. Once it tried to drive into a tree and once into a curb. I stopped it both times since i was in the driver's seat.

 

[mark95476]

I wonder if this was done on purpose.

If/when Tesla makes the "button" available I think we'll have all sorts of people posting videos of themselves crashing their vehicles to get the views and drive engagement. They see videos of people being irresponsible (like sleeping while driving) getting huge numbers....

And whoever was in the car filming couldn't stop it once they saw it headed towards a tree?

 

[Daniel in SD]

I wonder if this was done on purpose.

If/when Tesla makes the "button" available I think we'll have all sorts of people posting videos of themselves crashing their vehicles to get the views and drive engagement. They see videos of people being irresponsible (like sleeping while driving) getting huge numbers....

I think you're way overestimating YouTube revenue and way underestimating Tesla body shop prices. The most likely explanation is that people have not been watching YouTube videos of summon and are instead relying on the website description. Really.

 

[babbit]

For those asking why didn't he let go of the summon button, there's up to a 1 second delay from release to when the car will stop depending on how fast the car is going. Part of it is the lag time between your phone and the car's computer. The 2nd part is how long it physically takes the brakes to stop the car. Usually smart summon is slow enough where it can stop in a fraction of a second. However, I've seen it randomly accelerate with no rhyme or reason. In that situation, even letting go of the button probably would not have stopped the OP's car in this situation.

 

[qdeathstar]

He should of stopped pressing the button 5-10 seconds earlier tho

 

[AlanSubie4Life]

He should of stopped pressing the button 5-10 seconds earlier tho

Probably best to avoid pressing it at all.

 

[Daniel in SD]

Probably best to avoid pressing it at all.

Yep, there's even a toggle in the settings to prevent you from accidentally pressing the button. I'm not sure how Tesla can make it any safer.

 

[qdeathstar]

Probably best to avoid pressing it at all.

Well... can’t argue with that really. I only pressed it the one time.

 

[mark95476]

Big YouTubers make millions yearly and they all started small. I didn't watch and don't know anything about the poster.

I think you're way overestimating YouTube revenue and way underestimating Tesla body shop prices. The most likely explanation is that people have not been watching YouTube videos of summon and are instead relying on the website description. Really.

 

[DelPhonic1]

For those asking why didn't he let go of the summon button, there's up to a 1 second delay from release to when the car will stop depending on how fast the car is going. Part of it is the lag time between your phone and the car's computer. The 2nd part is how long it physically takes the brakes to stop the car. Usually smart summon is slow enough where it can stop in a fraction of a second. However, I've seen it randomly accelerate with no rhyme or reason. In that situation, even letting go of the button probably would not have stopped the OP's car in this situation.

1 second delay? Then don't let the car summon with that 1 sec delay taken into account. Personal responsibility.

 

[Daniel in SD]

1 second delay? Then don't let the car summon with that 1 sec delay taken into account. Personal responsibility.

Yet Tesla should have no responsibility for advertising a use case that violates that?

 

[linux-works]

you can't guess what the latency (jitter) is going to be like. you just cannot. to require that of wifi or BLE is, well, insane.

reminds me of a hack-a-day article where someone was proud of their stratum1 ntp server on an ESP chip. note, the esp chip has only 2.4ghz wifi and the timing accuracy of wifi is nowhere near as good as you'd need to even make use of that gps/pps receiver its attached to.

oh, and I'd also say that using a PHONE is the wrong thing for remote control, since that's yet another variable that will introduce huge amounts of jitter. if some background app does something or wants network i/o, oh boy, there goes your QoS.

just bad idea all around.