AutoPilot needs to look 2 cars ahead

It should also take into account what's going on in the adjacent lanes. I.e.
if both adjacent lanes start slowing down significantly it should probably increase follow distance automatically just to give a little more buffer if and when it's own lane starts slowing down. Other thing is autopilot should be able to recognize when the brake lights come on. If brake lights come on it should increase follow distance even if the car in front doesn't significantly slow down.

I've often thought much the same. If I see 3 lanes of bunched together cars ahead, all with brake lights on, I generally start slowing earlier than I might otherwise do when the car in front of me slows. This doesn't seem impossible for the AP to do as well.

I generally start slowing earlier than I might otherwise do when the car in front of me slows. This doesn't seem impossible for the AP to do as well.

Click to expand...

Agreed. I wish they read the break lights and take that also as an additional input, just like how humans would do.

As computers get more powerful, and software gets better, at least in theory AP can see further ahead than a human. Video cameras on the roof line and side mirrors will be able to see around traffic better than a driver can. When it can't see traffic ahead, it would adjust following distance and lane position accordingly. Humans will often move laterally in a lane to get a better a view, but it's limited by the fact we sit on one side of the car. Cameras have no such limitations. Lanes are typically 12' wide, so lane position and mirror mounted cameras could see around a large truck or motorhome, or the common SUV with limo tint on their back window.

Maybe they could use the charging snake tech to allow side cameras to look around and over cars ahead?

Cars are just going to have to communicate to each other.

There are too many times even human drivers get caught in situations where we simply can't see.

We shouldn't have to drive a big huge lifted SUV to see what's going on ahead of us.

kirkbauer said:

The single biggest limitation I see today is that AP doesn't seem to look further ahead than the car in front of you. This causes rougher driving than I would do in some cases.

It might not be possible to fully track 2 cars ahead but most of the time I think it would be with both the camera and radar. At the very least AP should be aware that there is a car right in front of the car in front of you even if it can't fully track it.

If the car 2 cars ahead starts slowing down then so should AP, or at least get ready to. More importantly, if the car in front leaves the lane and AP knows there is a car in front of that one, it should get a lock on that car before accelerating hard.

Click to expand...

I assume you mean TACC, which is only one part of AP. How do you propose that it should look through the car in front in order to see the one in front of that?

Nissan has been working on this for a while and has it on some of their 2016 cars. With their system, they're using the radar to look under the car in front of you. I wouldn't be surprised if the Tesla radar could do the same, based off of what it's already getting back. It's so low, there has to be some signal return from the car two ahead, but it's probably ignored by the programming.


Jump to 54 seconds.

Very soon the ability to see vehicles ahead of the one immediately in front (as well as behind and to the sides) will become much less important as all vehicles will be capable of communicating with all others so every one will know pretty much exactly where they all are relative to each other. Further more, when one needs to change speed or direction, all the other cars will instantly know and adapt their speed and direction to accommodate it. All this (and much, much more) is only a very few short years away.

It really does not take much imagination to see where all this will take us. For starters accidents will almost entirely disappear. Driving will be much less stressful as complete autonomy takes over. People imagine this is still decades away. In reality it will be mainstream in well less than 10 years (assuming no colossal global calamity befalls us). If you have any doubts about this, get yourself a ride in a Tesla Model S which already has about 70% full autonomous capability. I suspect the next version of Tesla AutoPilot hardware and software (already developed) will be capable of 100% autonomy even leaving out the sort of 'swarm communication' mentioned above.

MW

MW

I'm not even sure the system is currently paying attention to the speed of the car in front, instead of just the distance. That change, at least, could be done in software only and wouldn't require any new hardware. You should easily be able to get the speed of the car in front, and deduce a "they're braking really hard" type of events, even without recognizing the brake lights. On the other hand, if they could not only recognize the brake lights, but understand that some cars have the "more red" when they're braking hard, even better.

Their speed is much less important than your closing speed on them, which is exactly what the car is measuring with radar. I'd be surprised if it did any analysis on brake light status at all.

martinwinlow said:

Very soon the ability to see vehicles ahead of the one immediately in front (as well as behind and to the sides) will become much less important as all vehicles will be capable of communicating with all others so every one will know pretty much exactly where they all are relative to each other. Further more, when one needs to change speed or direction, all the other cars will instantly know and adapt their speed and direction to accommodate it. All this (and much, much more) is only a very few short years away.

Click to expand...

I wouldn't bet on that being only a "very few short years" away. There are a lot of old cars on the road, and there are a lot of people that won't trust their cars to drive themselves and will continue to drive manually even if they have a car that is capable.

commasign said:

It should also take into account what's going on in the adjacent lanes. I.e.
if both adjacent lanes start slowing down significantly it should probably increase follow distance automatically just to give a little more buffer if and when it's own lane starts slowing down.

Click to expand...

Careful: every commute I drive through an intersection with left and right filters.

cgiGuy said:

Their speed is much less important than your closing speed on them, which is exactly what the car is measuring with radar. I'd be surprised if it did any analysis on brake light status at all.

Click to expand...

If I see the car in front of me accelerating, even if I'm catching up with them, I react differently then if I see them slowing down. The software can do the same. So, not just look at your speed relative to them, but also the immediate history of your speed relative to them. If it's 20mph "now", but it was 30mph two second ago, that's a different story than if it's 20mph now, but it was 10mph two second ago.

MP3Mike said:

I wouldn't bet on that being only a "very few short years" away. There are a lot of old cars on the road, and there are a lot of people that won't trust their cars to drive themselves and will continue to drive manually even if they have a car that is capable.

Click to expand...

Exactly what I wanted to say. I'm just 33 but I'm not sure I will see this utopia.

martinwinlow said:

Very soon the ability to see vehicles ahead of the one immediately in front (as well as behind and to the sides) will become much less important as all vehicles will be capable of communicating with all others so every one will know pretty much exactly where they all are relative to each other. Further more, when one needs to change speed or direction, all the other cars will instantly know and adapt their speed and direction to accommodate it. All this (and much, much more) is only a very few short years away.

I suspect the next version of Tesla AutoPilot hardware and software (already developed) will be capable of 100% autonomy even leaving out the sort of 'swarm communication' mentioned above.

Click to expand...

I would say it will take very many long years for the auto industry to design, ratify, approve, and implement the "swarm communication" that you mention above. Even once that is done it will be many more years before a decent percentage of the cars have it.

As far as the "capable of communicating with all others so every one will know pretty much exactly where they all are relative to each other" comment, you certainly can't count on GPS accuracy for that, and I wouldn't think triangulating signals would be good enough either.

If you want to know where cars are your choices are laser/radar, which is the only way you are going to get 100% autonomy. You can't count on any swarm communication at all. Certainly you could add it to everything else you are doing, but is it worth the effort? (Since your system has to work 100% without it anyway.)

Whitmarsh said:

I assume you mean TACC, which is only one part of AP. How do you propose that it should look through the car in front in order to see the one in front of that?

Click to expand...

The same way humans do it (i.e. a mixture of looking through the glass and looking around the sides of the car in front).

mgboyes said:

The same way humans do it (i.e. a mixture of looking through the glass and looking around the sides of the car in front).

Click to expand...

Hardware, hardware, hardware. Not there yet.