About time to unveil the D and something else

[Ben W]

I had a safety question about the Autopilot changing lanes. Elon said you can push the turn signal and the car will change lanes for you. But what if another driver is coming up quickly from behind in your target lane? With only 16ft of surround visibility (ultrasonic), the car could merge straight into an unavoidable collision. I would think that changing lanes safely might require a longer-range backward view (radar), in addition to the ultrasonic and forward view.

 

[Yggdrasill]

I had a safety question about the Autopilot changing lanes. Elon said you can push the turn signal and the car will change lanes for you. But what if another driver is coming up quickly from behind in your target lane? With only 16ft of surround visibility (ultrasonic), the car could merge straight into an unavoidable collision. I would think that changing lanes safely might require a longer-range backward view (radar), in addition to the ultrasonic and forward view.

That occured to me as well. The car will need either a rearward facing camera or radar.

But for now, people will just have to check their mirrors before changing lanes. (How novel!)

 

[twinklejet]

to the above two posts, that's the idea behind why he's calling it autopilot and not autonomous. the driver makes the decision (after checking mirrors etc) that it is safe to change lanes, then signals for the lane change and ONLY THEN does the autopilot execute it after making a basic check that the 16 feet around the car is clear.. the autopilot still requires a pilot (driver) to give the command.. and if the driver fails to check for traffic before signalling to change lanes.. then darwinism will take over.

 

[Ben W]

to the above two posts, that's the idea behind why he's calling it autopilot and not autonomous. the driver makes the decision (after checking mirrors etc) that it is safe to change lanes, then signals for the lane change and ONLY THEN does the autopilot execute it after making a basic check that the 16 feet around the car is clear.. the autopilot still requires a pilot (driver) to give the command.. and if the driver fails to check for traffic before signalling to change lanes.. then darwinism will take over.

The "Autopilot vs autonomous" argument does not apply in this case, at least not in the usual way. That's because the turn signal has always only represented INTENT to merge; it has NEVER meant "It's safe to merge at this very moment"!! In fact, ordinary driving, it's most important to activate the turn signal precisely when it's NOT safe to merge, so that other drivers will see your intent and theoretically respond to allow you in. (When it's totally safe to merge, most drivers don't signal at all.) Asking people to change their deeply ingrained reflex of when to press the Turn signal, based solely on whether Autopilot is active, is just a recipe for disaster.

 

[Johan]

The "Autopilot vs autonomous" argument does not apply in this case, at least not in the usual way. That's because the turn signal has always only represented INTENT to merge; it has NEVER meant "It's safe to merge at this very moment"!! In fact, ordinary driving, it's most important to activate the turn signal precisely when it's NOT safe to merge, so that other drivers will see your intent and theoretically respond to allow you in. (When it's totally safe to merge, most drivers don't signal at all.) Asking people to change their deeply ingrained reflex of when to press the Turn signal, based solely on whether Autopilot is active, is just a recipe for disaster.

If that's how you use the turn signal, well, then you're doing it wrong.

 

[tomas]

If that's how you use the turn signal, well, then you're doing it wrong.

Hold on a sec. Ben is 100% correct. Turn signal is not to tell people you are turning but that you intend to turn. Back to basic driver ed, it should be activated well in advance of turn or lane change. If it activates the lane change immediately that is not only a behavior change for good drivers, but also improper use of signal.

 

[Johan]

Hold on a sec. Ben is 100% correct. Turn signal is not to tell people you are turning but that you intend to turn. Back to basic driver ed, it should be activated well in advance of turn or lane change. If it activates the lane change immediately that is not only a behavior change for good drivers, but also improper use of signal.

... Yes BUT it is also directly wrong to activate the turn signal before having checked that the conditions allow you to turn/change lanes since that can create dangerous situations.

I agree though that it should be a two step process. Perhaps two "clicks" on the stalk: 1st nudge: blink, 2nd (further) nudge: keep blinking and change lane. If all clear: nudge all the way to the second position immidiately.

Edit: it's probably like this already: nudge stalk without making it click in to position: blink 3 times but don't change lane. Click on position: blink and change lane automatically. So there's probably no problem here.

 

[Ben W]

If that's how you use the turn signal, well, then you're doing it wrong.

To be clear, I'm not excusing drivers who never signal. My point is that activating the turn signal, and physically merging, are two independent actions, and the Model S Autopilot conflates them in a potentially dangerous way.

According to the CA DMV, turn signals should be activated at least five seconds before changing lanes on the freeway. (source: California Driver Handbook) If the Model S Autopilot ignores this guideline, and effectively merges without prior signaling, then that's arguably illegal, and certainly hazardous to other drivers. (It would make me quite leery of passing an autopiloted Tesla on the left, for instance.) But if the Model S Autopilot DOES follow this guideline, then you have a situation where the driver presses the turn signal, and at some vague indeterminate time in the future (because brains are not stopwatches) the car will effectively begin merging by itself. The driver will not be able to anticipate this precise moment, so his reaction time will be slower (as far as re-checking for dangers) than if he controlled the lane change directly, and that's why it's crucial for the autopilot to detect these potential collisions on its own. With e.g. Adaptive Cruise Control this is less of an issue, because the driver is always at least looking in the direction of the potential collision.

 

[Ben W]

... Yes BUT it is also directly wrong to activate the turn signal before having checked that the conditions allow you to turn/change lanes since that can create dangerous situations.

In heavy traffic (e.g. the 405 at rush hour), signaling one's intent well in advance, and hoping for a kind-hearted driver to see your turn signal and make space for you, is often the ONLY way you can squeeze your way into a very crowded lane.

In my own driving in thick traffic, I try to anticipate a reasonable merge opportunity and put on my turn signal a few seconds in advance, to let the cars around me know what I intend to do. In practice I'm often nearly parallel with the car in front of the gap when I start signaling; then I slow down a bit and merge in. (This is so that the car I'm cutting on has a clear view of my turn signal, and hopefully will slow down themselves and let me in.) The delay between signaling and actually merging is very intentional.

 

[clea]

Yes, yes, of course he meant blowing snow and such and it was in the context of seeing objects. That doesn't mean because a speed sign is covered by snow, or the lines on the road are covered by snow that suddenly the car's autopilot features cease to work. There are four systems that work together and/or back each other up. In the example given, if a sign is covered by snow then the car will logically revert to nav/Google/Garmin/fill in the blank info.

From a purely realistic view, having driven in snowstorms, the posted speed of a road is of zero relevance. If you want to stay out of the ditch, you aren't driving the speed limit, let alone over it. You drive at a speed dictated by the weather conditions. Posted speed limits are obviously for ideal driving and road conditions. This is not autonomous driving, it's autopilot which requires human supervision. It will continue to advance as the months and years go by, such that those things that aren't yet worked out will get worked out.

This brings up a question I have and didn't see it covered yet ... Is the regen going to be coming from the front drive as well on the D models? This could possibly be an excellent means of assisting the braking on ice as opposed to having the TC kick in on the rear wheels only during regen on slippery surfaces.

 

[scaesare]

This brings up a question I have and didn't see it covered yet ... Is the regen going to be coming from the front drive as well on the D models? This could possibly be an excellent means of assisting the braking on ice as opposed to having the TC kick in on the rear wheels only during regen on slippery surfaces.

Given that weight transfers forward during deceleration, I hope so... that would seem to offer opportunity for more braking without loss of traction.

 

[Saghost]

That occured to me as well. The car will need either a rearward facing camera or radar.

But for now, people will just have to check their mirrors before changing lanes. (How novel!)

The Model S has always had a standard high definition rearward facing camera that can see straight back as well as down unless I'm completely confused. The backup camera wasn't mentioned in the autopilot pitch, but there's no reason the car couldn't be doing object recognition against it as well to address the fast mover concern. (And no evidence that the car is, so it's just speculation from me so far.)
Walter

 

[pete8314]

I've read this entire thread since it started, and one omission seems to be (unless I missed it) the timescales for the cool new auto-pilot features. I know the lane assist and speed assist are functional, but is anyone concerned, based on track-history, that the uber-cool features, such as self park, self-pickup etc are months, or years away? There wasn't even a demo of any of the parking features, so it sure seems more like vaporware at the moment, and my fear, much like the other vaporware (anything from battery swapping, to using the internal HDD for music storage), is it'll never appear, or take years, and possibly even more hardware once they're really fine-tuned the mechanics of it.

Don't get me wrong, it looks, and sounds amazing, but with no commitment to timescales, it does seem a bit vague.

Edit: I guess it's being discussed in this theread: Schedule for Autopilot features

 

[Krugerrand]

This brings up a question I have and didn't see it covered yet ... Is the regen going to be coming from the front drive as well on the D models? This could possibly be an excellent means of assisting the braking on ice as opposed to having the TC kick in on the rear wheels only during regen on slippery surfaces.

Elon said the car would be able to adjust breaking and acceleration between the front and back motors instantaneously for handling and traction (AWD) purposes.

As for the whole change lane discussion...*le sigh, people, le sigh* Firstly, the car can see 360 degrees around it. It will see that vehicle coming from behind and I imagine it'll know how fast it's coming, otherwise what would be the point as not all objects around the car will be stationary or moving. If it turns out that 16' is not enough distance/time for the autopilot features to work then I make the educated guess that Tesla would change it to be 20' or a million feet, whatever. I would also make the educated guess that the car will NOT change lanes simply because a driver puts on the turn signal, but that the code is written something (a lot more involved) like this

10 left turn signal indicator on
20 Sensors 1 thru whatever check around the car from directly in front, along left side of vehicle, and to back of vehicle for stationary and moving objects
30 If one sensor reports back with object detection car does NOT change lanes - go to line 20
40 If all sensors report back the all clear then car changes lane


Before you panic and think the world is coming to an end, try and remember that autopilot and autonomous driving are works in progress and that no one is forcing you (at this moment in history) to use the autopilot setting on the car. Furthermore, Elon specifically said that autopilot still requires driver responsibility. He also specifically said that the driver could manually override the autopilot. It would be difficult, but it could be done. Pilots in airplanes do not set the plane on autopilot and then decide to take a nap. They remain in their seats to continue to monitor all systems of the plane.

 

[Yggdrasill]

The Model S has always had a standard high definition rearward facing camera that can see straight back as well as down unless I'm completely confused. The backup camera wasn't mentioned in the autopilot pitch, but there's no reason the car couldn't be doing object recognition against it as well to address the fast mover concern. (And no evidence that the car is, so it's just speculation from me so far.)
Walter

That's true. Didin't even occur to me that they could possibly integrate that camera into the autopilot. If they do this, the sensor coverage should be sufficient, I think.

 

[Krugerrand]

That's true. Didin't even occur to me that they could possibly integrate that camera into the autopilot. If they do this, the sensor coverage should be sufficient, I think.

Specifically stated and diagrams show, the car sees 360 degrees.

 

[Saghost]

That's true. Didin't even occur to me that they could possibly integrate that camera into the autopilot. If they do this, the sensor coverage should be sufficient, I think.

I really wish they'd add side/down looking cameras to the mirror housings. These could enable a "birdseye" view like some other cars do in the short term for parking help - and would be the final set of sensors the car needs for full autonomy in the long term. As it is I don't think the car has any way of checking left and right at a stop sign or traffic light. (Not that it'll be doing that this year anyway.)
Walter

- - - Updated - - -

Specifically stated and diagrams show, the car sees 360 degrees.

I'm pretty sure that's a reference to the ultrasound, which is very short range.
Walter

 

[Yggdrasill]

As for the whole change lane discussion...*le sigh, people, le sigh* Firstly, the car can see 360 degrees around it. It will see that vehicle coming from behind and I imagine it'll know how fast it's coming, otherwise what would be the point as not all objects around the car will be stationary or moving. If it turns out that 16' is not enough distance/time for the autopilot features to work then I make the educated guess that Tesla would change it to be 20' or a million feet, whatever.

16 feet isn't enough. On the autobahn, you might be driving at 65 mph in the left lane, with cars passing in the right lane at 150 mph. If you see that car when it's 16 feet away, you have absolutely no chance of getting out of the way. With a difference in velocity of 85 mph, those 16 feet will be covered in 0.13 seconds.

The parking sensors don't work at distances much greater than 16 feet. That's why you have the radar in the first place.

 

[Krugerrand]

I'm pretty sure that's a reference to the ultrasound, which is very short range.
Walter

Maybe so, but what does that matter? Autopilot is a WIP. If it needs to change down the road, it'll be changed. Has Tesla not already shown a willingness to make changes when one is warranted?

People are getting way ahead of themselves and hand-wringing over things that are at present irrelevant.

- - - Updated - - -

16 feet isn't enough.

Are you under the impression that what was presented on Thursday is the final iteration? If 16 feet isn't enough, then Tesla will change it when the situation calls for it.

Since you think you know better what Tesla should be doing at this phase of the autopilot development program, perhaps you should apply for a job?

 

[Beavis]

How long before we see true autonomous? The D has autopilot but it looks like it is only intended for highway use to keep you in a lane as opposed to being able to navigate city streets. On the other hand, you can summons the car to you if you are on private property. Does the D have all the autonomous hardware and is just awaiting a software upgrade?