Elon: "Feature complete for full self driving this year"

[ramonneke]

Why don't traffic lights emit radio signals to tell cars what the traffic light colour is.

We have such a system here in The Netherlands. Traffic lights report via internet:

- Current state (red, green, amber)
- If the light will go green immediately if the crossing has no other approaching cars from other directions
- Time until green
- Recommended speed to keep traveling to have a "green wave".

Its under test and likely will expand to more trafficlights in the future.

 

[ramonneke]

So Tesla is releasing a feature that requires active driver supervision, but is not providing proper driver monitoring (no driver facing camera) and saying "oops the feature may fail but you need to pay attention". That is terrible! Tesla is being incredibly irresponsible with this feature.

Its not like the car will brake hard. It will coast and slow down gently like you normally would do when approaching a crossing (I hope, even with green..)

 

[Daniel in SD]

So, maps are a crutch that Tesla plans to get rid of in the future?
I’m sure every AV company would get rid of maps if they could achieve the required level of safety without them. It costs money to maintain the maps. I doubt they ever will because it doesn’t cost very much money.

I'm going to make the bold prediction that if Tesla ever releases stop sign and stop light response they will use maps (or they're going to have an absurdly high error rate).

 

[willow_hiller]

Interesting edge case: the car phantom braked because it mistook red lights on the back of a truck as a bunch of traffic lights. Derek Teay on Twitter

This is obviously incorrect behavior from the system, but doesn't this also debunk the idea that the vision system needs a map in order to recognize a stoplight?

Unless their map happened to have hundreds of lights mapped in the middle of the highway.

 

[mongo]

You keep changing the goal post. Stick to one thing. The traffic light/sign stop system DOES NOT work without a map PERIOD.

There is no goal post changing. The current system is gating on known intersections, that does not mean the design intent is to gate on known intersections.

Now you are again changing the goal post and EVEN then the traffic light/sign stop system NEEDS the Map.
It doesn't work without it, ITS A REQUIREMENT. As pointed out above and you just dismissed.

You yourself posted that the vehicle detects unmapped light. Therefore, it does not NEED it to be mapped to detect it. The decision to not act on unmapped lights in the current release is due to it being in development.

You previously said Tesla weren't using maps. Now all of a sudden its okay?

When did I say Tesla didn't currently use maps? If I did, that was my error.

Again you and others argued for months that using maps for anything other than routes were a clutch..

Yeah, and Tesla is using a crutch to help develop the system in a safer way. The question is: are they moving toward a system when the maps are not required?

Moreover this has already BEEN proven. You disregard it just like anything. This whole discussion is pointless because you keep refusing all facts and evidence.

A newly developed Traffic light that is NOT in the map is detected using vision but is ignored because its not in the map. So no control algorithm is initiated.

"I did test 2020.12.1 with new map, never panic with AP on while trying to hit new traffic light recently developed. Tried it 10 times and the car can see the traffic light but never gave me alert about it, while it always gave me alerts about old stop sign and traffic lights"
Mike Alani on Twitter

Another Mapped traffic light that conflicts with vision
Victor on Twitter

Ah, misread that first tweet, though it was was testing 2012.12.1 in general and not getting stopping action, not new light vs old light. Supports that Tesla is gating the vision based on maps currently. Also shows that in this location, they do not need to.

In the second example, I would have thought that was a red light.

 

[diplomat33]

OK, but does Waymo have a plan / goal to ever not require maps?

I don't know Waymo's plans. I assume they will continue to use maps. Why are you assuming they need a plan to stop using maps? Maps are good for autonomous cars if you want them to be safe and reliable.

I feel like the debate over HD maps is similar to the debate over lidar.

Would it be ideal to have an autonomous car that is so intelligent and capable that it can drive autonomously safely and reliably with just a few cameras and no HD maps because it can interpret any new environment and do the right thing the first time? Sure. But we are not there yet. So right now, we need things like lidar and HD maps to give autonomous cars that little extra reliability needed if we want to deploy safe and reliable robotaxis with no safety drivers.

 

[diplomat33]

This is obviously incorrect behavior from the system, but doesn't this also debunk the idea that the vision system needs a map in order to recognize a stoplight?

Unless their map happened to have hundreds of lights mapped in the middle of the highway.

Again, the issue is not whether or not camera vision can recognize a stop light. Obviously, you can train camera vision to recognize traffic lights. The issue is reliably. If anything, this proves why you need maps. Without maps, you are just relying on camera vision and it won't be very reliable because your camera vision will make mistakes like this one. Sure, you can take the Tesla approach of just taking each edge case and retraining the NN each time but there are a ton of edge cases. That takes time and your system will continue to be unreliable for awhile. With maps, the car would have known to dismiss this case since the maps would have indicated no traffic light in that location. So maps help solve cases where your camera vision thinks it sees a traffic light but it's wrong.

 

[Bladerskb]

This is obviously incorrect behavior from the system, but doesn't this also debunk the idea that the vision system needs a map in order to recognize a stoplight?

Unless their map happened to have hundreds of lights mapped in the middle of the highway.

No this is simply car phantom braking. remember the update for actually braking is NOT out yet. The only thing cars do now is give the red light warning and you don't see the red light warning in that video.

 

[Bladerskb]

There is no goal post changing. The current system is gating on known intersections, that does not mean the design intent is to gate on known intersections.

Yes you are. Design intent means absolutely nothing. Because for example, in 2024, Let say Tesla has a autonomous system that works in California but not other locales. You will still be saying that its general and not brittle because the design intent is eventually for everywhere in some unknown future date. You could literally try to use that excuse to scapegoat indefinitely all the way to 2030 and beyond.

This is typical. Always looking at the future and disregarding and refusing everything that has happen in the past and in the present. Then next year, when the future inevitably becomes the present you therefore ignore it and create a new future is created to continue the narrative and the beat goes on.

You yourself posted that the vehicle detects unmapped light. Therefore, it does not NEED it to be mapped to detect it. The decision to not act on unmapped lights in the current release is due to it being in development.

Here you are again changing the goal post.

Detection was never the contention. No one said you need map to detect something. No one.
This was understood even by you just acouple months ago. Now you are literally making stuff up as you go. Just admit you were wrong. Period. I admit i'm wrong when i say something. Why is that so hard so you?

You months ago:

If you ignore your vision data because it doesn't match your HD map, then the map is the primary data source.

Its perception VS HD Map. But now you are twisting it. You are literally claiming that Waymo cars are incapable of detecting lanes without their Map. Its absolute none sense. I mean come on, just admit you are wrong. Its okay i have been wrong dozens of times.

The decision to not act on unmapped lights in the current release is due to it being in development.

Not acting on un-mapped lights is literally what we are talking about. Its literally what you just said isn't happening.
You are insufferable. I dont know how @diplomat33 does it.

Not sure how you came to your conclusions about how the system operates. If there is a known intesection: it stops. If there is a traffic control device detected, it stops. Stopping where there isn't a light protects against the cases of pedestrians and impropper cross traffic.

I think maybe you were referring to this warning:
"Yield signs or temporary traffic lights or stop signs (such as at construction areas)"

As in, atypical arrangements NOT in terms of unmapped intersections. Big difference, huge.

 

[Eno Deb]

Interesting edge case: the car phantom braked because it mistook red lights on the back of a truck as a bunch of traffic lights. Derek Teay on Twitter

This is not necessarily triggered by the rear lights. I get phantom braking with trucks in the neighboring lane quite often, and I'm on HW2.5 ...

 

[diplomat33]

This is not necessarily triggered by the rear lights. I get phantom braking with trucks in the neighboring lane quite often, and I'm on HW2.5 ...

True. I probably should have rephrased my post.

There are probably 2 separate issues that happen to be going on at the same time in the video:
1) Phantom braking because of the truck but not because of the lights.
2) The visualization is wrong because it thinks the red lights on the trucks are traffic lights.

But #2 does point to an edge case that Tesla needs to solve since the car is clearly confused by the lights on the truck.

 

[Eno Deb]

So, from the leaked manual it looks like the car will always stop at intersections unless the driver takes action, even if the light is green? That sounds like it will invite angry honks (in the best case) or rear-ending accidents if the driver isn't constantly watching the screen.

 

[diplomat33]

So, from the leaked manual it looks like the car will always stop at intersections unless the driver takes action, even if the light is green? That sounds like it will invite angry honks (in the best case) or rear-ending accidents if the driver isn't constantly watching the screen.

Yep. I already mentioned this problem.

 

[Eno Deb]

From a usability and safety perspective I think this is all going in the wrong direction, just like NoA. In addition to monitoring your surroundings, you are additionally also expected to constantly keep an eye on the screen to see what your own car is up to. Sensory load and stress is increasing rather than decreasing.

 

[linux-works]

I think I now understand why tesla in overloading the accel pedal for an ACK or confirm.

I have major concerns about the quality of the stalks and I'm sure I'm not alone. they are not something you can bet your life on, and yet, they are now being asked to do things that control important vehicle status. if they were the ONLY way you could confirm that you want to go thru a green light (sigh) and if the stalk broke or malfunctioned, that would be a huge safety issue. so, I can see that the accel pedal is probably more robust and more reliable. not sure if it has redundant sensors but it should.

at this point, I'd want the stalk to have redundant sensors, as well. maybe an electrical switch and an optical one or hall effect. 2 diff techs that would be sampled to determine if the stalk really was pushed or not.

I could train myself to do lots of 'stalk' things but again, the thing is going to wear out and I don't think they built it as well as they should, given how much more use its now going to see.

 

[Daniel in SD]

I can see that the accel pedal is probably more robust and more reliable. not sure if it has redundant sensors but it should.

It does but when the sensor disagree it just turns off. Personally I think worrying about mechanical failures is silly. Cars fail all the time, and sure a certain percentage of the time it causes an accident, but not being able to accelerate is a pretty safe failure mode.

From a usability and safety perspective I think this is all going in the wrong direction, just like NoA. In addition to monitoring your surroundings, you are additionally also expected to constantly keep an eye on the screen to see what your own car is up to. Sensory load and stress is increasing rather than decreasing.

I don't use NoA or Smart Summon for the same reason, it's easier and less stressful to do it myself. Some people are going to love this feature, that's fine. I don't think people will have to look at the screen, I bet they'll get a feel for when the car is going to start stopping and press the accelerator before it starts slowing down. I'd also imagine that it will start slowing down very slowly at first, it's probably perceptible to the driver way before it would cause a problem for someone following.

 

[emmz0r]

Seems like another thing that will work half the time and add to stress instead of convinience.

The GPS based speed limits are bad enough.

 

[linux-works]

It does but when the sensor disagree it just turns off. Personally I think worrying about mechanical failures is silly. Cars fail all the time, and sure a certain percentage of the time it causes an accident, but not being able to accelerate is a pretty safe failure mode.

counterpoint: some tesla is barreling in, behind me, and does not look like they are planning on stopping. maybe I look like a firetruck (hmm, "too soon?"). at any rate, I may NEED to get the hell going, to avoid a problem.

I see accel almost as valuable as braking and steering, to get out of problem situations.

besides that, I'm real big on redundancy when you can. its cheap in the long run and can save lives. diversity for sensor and actuator types adds cost but I personally think its worth it. bean counters disagree, but end-users never say they want less reliability in anything.

if the stalk is going to become a very key part of driving, actively, it better be closer to asil-d than not.

 

[linux-works]

So, from the leaked manual it looks like the car will always stop at intersections unless the driver takes action, even if the light is green? That sounds like it will invite angry honks (in the best case) or rear-ending accidents if the driver isn't constantly watching the screen.

its going to hurt the brand image, that is for sure.

I wonder if us, being beta subjects, makes that worth it to tesla? we are doing a lot more manual work to teach their network. if I'm going to wear out my stalk and tire out my right arm more, I want a serious discount on FSD upgrade, man (half serious).

 

[Daniel in SD]

I see accel almost as valuable as braking and steering, to get out of problem situations.

You're talking about an incredibly unlikely event. Your accelerator pedal would have to fail at exactly the same time as you need to accelerate. It's not going to happen.
The only systems that I'm aware of on a car that are redundant are, the hood release, the master cylinder, and the accelerator pedal sensor.