Firmware 9 in August will start rolling out full self-driving features!!!

[rnortman]

The actual patent has a number of different cases. At least one could be applied to auto overtaking:

And this one looks like it is a generic catch-all for self-driving:

I don't think either of those apply to self-driving. In both cases the system is trying to figure out if the car intends to leave the lane based on sensor inputs. (Specifically: "based on data from at least one of the camera or steering angle sensor".) A self-driving system knows its own intentions without relying on sensor inputs.

Plus there is plenty of prior art for self-driving cars turning on their own turn signals. Waymo cars have been doing this for years now, among others. Tesla would be laughed out of court if they claimed this applied to an autonomous vehicle turning on its turn signal. This is entirely about figuring out what the human driver intends to do and turning on the signal for them -- turning it on too late really, but better late than never? So this is for competing with other automakers developing L2 (or even L1) driver assistance features.

 

[J1mbo]

(Specifically: "based on data from at least one of the camera or steering angle sensor".)

Camera sensor data needs to be interpreted by the NN.

 

[rnortman]

Camera sensor data needs to be interpreted by the NN.

Yes, and? How specifically does this patent have anything to do with self-driving or auto lane change?

 

[J1mbo]

Yes, and? How specifically does this patent have anything to do with self-driving or auto lane change?

Really?

[0011] An embodiment includes an apparatus for automatic turn signal source activation, the apparatus comprising: a vehicle including a processor, a camera, a steering angle sensor, and a turn signal source, wherein the processor is programmed to determine when the vehicle is going to turn or leave a lane based on data from at least one of the camera or the steering angle sensor, wherein the processor is programmed to activate the turn signal source when the processor determines that the vehicle is going to turn or leave the lane.

I would guess that the "prior art" all relies on HD mapping to do this.

 

[SlicedBr3ad]

So many hopes ruined for end of August v9

 

[R.S]

Really?



I would guess that the "prior art" all relies on HD mapping to do this.

This is for making the car use a turn signal, when the driver forgets, this is a simple assistance and not needed for FSD.

An autonomous car knows when it will change lanes, it does it willingly. This is for all the people that don’t use turn signals.

But it seems pretty useless. The only interesting part IMO is the GPS based turn signal. That way navigation could be more practical, because your car will use the turn signals accordingly.

Otherwise the turn signal will only result in false positives, or be super late.

BACKGROUND

[0002] A vehicle, such as a car, hosts a turn signal light source, such as a bulb, which is manually activated by a driver of the vehicle, such as when making a turn or switching lanes. When activated, the turn signal light source visually notifies others, such pedestrians and the drivers of other vehicles, that the vehicle may change its direction of travel by, for example, turning onto another road or switching road lanes. However, since the turn signal light source is manually activated by the driver, there are often times when the driver neglects to activate the turn signal light source before changing its direction of travel. This situation can be dangerous, especially on high speed roads, and is often a cause of vehicle accidents or instances of road rage.

[0003] Although various technologies exist in order to mitigate the failure of a driver to engage the vehicle's turn signal, these technologies are inadequate for various technical reasons, such as false positives and slow reaction time. For example, some prior attempts to ameliorate this situation have included smart turn signals that stay silent and in the background, but present a warning to the driver if the vehicle is steered outside the bounds of a lane, and turn signal assist programs that generate dashboard messages to remind a driver to use turn signals after repeated failures to do so. These existing technologies still rely on the driver of the vehicle to manually engage the turn signal light source when appropriate.

 

[R.S]

The only question is does he mean August 2018?

Obviously not, duh!

Edit: it might seem that this post was written way later, than it really was. That’s just Elon time encroaching into the forum.

 

[rnortman]

Obviously not, duh!

Edit: it might seem that this post was written way later, than it really was. That’s just Elon time encroaching into the forum.

Hey, don't give up too soon; they still have all day to get this done.

 

[rnortman]

I would guess that the "prior art" all relies on HD mapping to do this.

What does HD mapping have to do with anything? You also have various companies other than Tesla trying to do autonomous driving without maps, and academic/military R&D efforts have been doing it going back decades, and I guarantee you that all of them signal lane changes before they start to execute the maneuver. Having an autonomous vehicle signal its intent to turn is simply not patentable, and it is not what this patent is about. And anyway, map-based AVs also need to process sensor input to figure out where they are on the map and where other vehicles are. By your reading of the patent, this would apply to them. Oh sugar, I bet Waymo is quaking in their boots after reading this!

You know what is common between map-based AVs and Teslas? Neither of them need sensor input to know that they're about to turn. They need sensor input to decide to turn. You only need sensor input to know you're about to turn if you're not the one making the decision -- i.e., when the human is doing the steering.

 

[J1mbo]

What does HD mapping have to do with anything? You also have various companies other than Tesla trying to do autonomous driving without maps, and academic/military R&D efforts have been doing it going back decades, and I guarantee you that all of them signal lane changes before they start to execute the maneuver. Having an autonomous vehicle signal its intent to turn is simply not patentable, and it is not what this patent is about. And anyway, map-based AVs also need to process sensor input to figure out where they are on the map and where other vehicles are. By your reading of the patent, this would apply to them. Oh sugar, I bet Waymo is quaking in their boots after reading this!

You know what is common between map-based AVs and Teslas? Neither of them need sensor input to know that they're about to turn. They need sensor input to decide to turn. You only need sensor input to know you're about to turn if you're not the one making the decision -- i.e., when the human is doing the steering.

HD mapping could include pre-defined signalling zones around junctions, so that the car does not need to make the determination itself. It just reads a "signal left" flag from the map data and actions it until the turn is complete.

The car needs sensor input from the camera(s) to decide when it can overtake. Once the car has made that decision autonomously, it can start to indicate as the first step of the overtake. This is covered in the section I quoted.

Applying the "this one is only for lazy humans" principal makes the whole exercise completely pointless. The auto-indicator will always be trying to second-guess the driver. It will not start until the car is halfway through a lane change; or after the driver has commited to a left or right turn at a crossroad. It will be dangerous on roundabouts, because it will not indicate that you are going to exit until you have started to exit.

It is clear that Tesla have gone for the broadest possible application of the principal that the car can decide when to activate a lighted turn signal. Nobody is saying it is bulletproof. Tesla have open-sourced all their patents, so Waymo et al can relax.

 

[R.S]

HD mapping could include pre-defined signalling zones around junctions, so that the car does not need to make the determination itself. It just reads a "signal left" flag from the map data and actions it until the turn is complete.

The car needs sensor input from the camera(s) to decide when it can overtake. Once the car has made that decision autonomously, it can start to indicate as the first step of the overtake. This is covered in the section I quoted.

Applying the "this one is only for lazy humans" principal makes the whole exercise completely pointless. The auto-indicator will always be trying to second-guess the driver. It will not start until the car is halfway through a lane change; or after the driver has commited to a left or right turn at a crossroad. It will be dangerous on roundabouts, because it will not indicate that you are going to exit until you have started to exit.

It is clear that Tesla have gone for the broadest possible application of the principal that the car can decide when to activate a lighted turn signal. Nobody is saying it is bulletproof. Tesla have open-sourced all their patents, so Waymo et al can relax.

You are thinking of it the wrong way. If the car changes a lane based on Autopilot input, then it doesn't need additional camera and steering sensor input to do so.

The important section is this:

[...] wherein the processor is programmed to activate the turn signal source when the processor determines that the vehicle is going to turn or leave the lane.

The processor doesn't determine that the car can leave the lane, or that it should turn, but rather that it already does.
This patent is just for automatic use of turn signals, based on a human driving the car.

 

[rnortman]

Applying the "this one is only for lazy humans" principal makes the whole exercise completely pointless. The auto-indicator will always be trying to second-guess the driver. It will not start until the car is halfway through a lane change; or after the driver has commited to a left or right turn at a crossroad. It will be dangerous on roundabouts, because it will not indicate that you are going to exit until you have started to exit.

On all of this I agree. On everything else, you're still missing the point and need to re-read the patent. Nobody needs a patent for turning on a turn signal. The patent is for detecting that the vehicle is leaving its lane.

Now the really interesting thing, now that I think about it, is that Teslas and a bunch of other cars already do this in the form of Lane Departure Warning systems. The thing Tesla is adding is that if you drift out of your lane and no other cars are in that lane, instead of activating the lane departure warning, it turns on your turn signal for you. Personally, I'd rather have it activate the land departure warning; this seems dangerous on top of being pointless. (Or maybe it turns on the turn signal *and* alerts you with the rumbling steering wheel? I guess they have that option.)

Here would be a useful feature, if they could make it work: Beef up the lane departure warning so that it actually *prevents* you from making the lane change if there's a car there. But that's really really hard to implement because you'd need the false positives to be very nearly zero or else it will be unsafe -- it might cause an accident because of a false positive rather than preventing one. Or maybe you need to sideswipe that car to avoid a head-on collision with a different vehicle that just crossed the median; I think that would be the right thing to do in most circumstances and I wouldn't want my car to stop me from doing it.

Anyway, pointless patent. Nobody wants this feature.

 

[MP3Mike]

Hey, don't give up too soon; they still have all day to get this done.

If everybody is doing their jobs correctly, and holding up their end of the contracts they signed, nobody would know if V9 has hit the EAP cars yet or not...

 

[Dutchie]

So many hopes ruined for end of August v9

Still a couple of hours left. I predict a massive roll-out starting this afternoon!

 

[Kant.Ing]

When a patent is published, it's already a prior art. Now they should be doing behavior prediction for other drivers/cars or spending processor bandwidth on detecting others' blinkers.

 

[Zooomer]

Maybe this is where Tesla responds with 'V9 is ready but we want to make absolutely sure every piece of it is safe and validated. At Tesla safety is our #1 priority'.

Then everyone here starts saying 'quit complaining, it's a good thing Elon wants us to have safe cars'. As if somehow you can never be upset for simple trying to hold someone to their word. Remember, it was only Aug 1st that Elon said 'in about 4 weeks' which was technically up 3 days ago. (insert guys saying 'about' means anything).

 

[R.S]

When a patent is published, it's already a prior art. Now they should be doing behavior prediction for other drivers/cars or spending processor bandwidth on detecting others' blinkers.

That’s not true, most parents containing SW get published before they are fully implemented. Otherwise someone else could patent it before you do.

I have had some HW patents we patented before they were technically feasible to mass produce.

On all of this I agree. On everything else, you're still missing the point and need to re-read the patent. Nobody needs a patent for turning on a turn signal. The patent is for detecting that the vehicle is leaving its lane.

Now the really interesting thing, now that I think about it, is that Teslas and a bunch of other cars already do this in the form of Lane Departure Warning systems. The thing Tesla is adding is that if you drift out of your lane and no other cars are in that lane, instead of activating the lane departure warning, it turns on your turn signal for you. Personally, I'd rather have it activate the land departure warning; this seems dangerous on top of being pointless. (Or maybe it turns on the turn signal *and* alerts you with the rumbling steering wheel? I guess they have that option.)

Here would be a useful feature, if they could make it work: Beef up the lane departure warning so that it actually *prevents* you from making the lane change if there's a car there. But that's really really hard to implement because you'd need the false positives to be very nearly zero or else it will be unsafe -- it might cause an accident because of a false positive rather than preventing one. Or maybe you need to sideswipe that car to avoid a head-on collision with a different vehicle that just crossed the median; I think that would be the right thing to do in most circumstances and I wouldn't want my car to stop me from doing it.

Anyway, pointless patent. Nobody wants this feature.

I disagree, if it’s an addition to lane departure warning I think it could add safety.

And if it’s GPS based it could also be practical, if you are in a city you don’t know for example. In situations like that I sometimes get so confused by where to go, that I indicate too late.

So not pointless, but not as breathtaking as @J1mbo thinks it is.

 

[Kant.Ing]

That’s not true, most parents containing SW get published before they are fully implemented. Otherwise someone else could patent it before you do.
I have had some HW patents we patented before they were technically feasible to mass produce.

The scope is inside Tesla, not other competition. My point is that at the time Tesla's patent gets granted through legal process, the technical team might have come up with something better already, either in concept or implementation.

 

[rnortman]

When a patent is published, it's already a prior art. Now they should be doing behavior prediction for other drivers/cars or spending processor bandwidth on detecting others' blinkers.

This is not what "prior art" means. A patent can be prior art with respect to other patents, but not with respect to itself.

Prior art - Wikipedia

These patent posts are turning into word salad. I honestly don't know what position anybody is arguing anymore.

 

[Pollux]

There's probably an interesting metric that could be devised: number of foreheads slapped, and at what level of strength, all throughout Tesla whenever Elon Musk puts a date on something.

Alan