6.2 TACC still following WAY too far behind

[wk057]

Why not make enough range of settings to keep everyone happy?

• There are only 7 physical clicks on the stalk....
  • ... and most people seem happy with the range currently available
• Someone accidentally using the follow-at-6-inches setting and getting in an accident would sue Tesla
  • Having a setting that makes the following distance unsafe, reasons stated above, would be unwise in this regard, especially given Tesla's mission to bring about semi-autonomous/autonomous vehicles in the future
• A setting much higher than existing 7 would probably be less useful on many roads if the vehicle ahead could be out of line of sight long enough for the system to think it's in the clear, accelerate, only to find that the other vehicle again a moment later and slow down
  • Hills, curves, etc
• Many others.

 

[billarnett]

Speaking of curves, wouldn't it be a good idea for TACC to postpone accelerating when it loses contact with the car in front if the steering angle is currently greater than some small amount? It should be ok to accelerate in a freeway corner but not much more. This wouldn't fix the whole problem (the car I. Front may disappear before you reach the corner) but might help a bit.

Also, I don't quite like the way TACC interacts with the overall set speed. I seem to usually have to set it to a higher speed than I really like just to account for the variations in the other guy's speed. Might it not be a good idea to let TACC go 5 mph or so over the set speed when it's in contact with the car in front? That way I could let the set speed be more like what I really want the average to be.

 

[NoMoGas]

Speaking of curves, wouldn't it be a good idea for TACC to postpone accelerating when it loses contact with the car in front if the steering angle is currently greater than some small amount? It should be ok to accelerate in a freeway corner but not much more. This wouldn't fix the whole problem (the car I. Front may disappear before you reach the corner) but might help a bit

it already does, could maybe use a tweak but if you're on a sharp curve it won't accelerate until you straighten out.

 

[wk057]

I was more referring to when you weren't in the curve and the person in front was (in reference to a > 7 setting). But yeah, it already stops acceleration when in a turn/curve of any significance.

 

[billarnett]

is that new in 6.2? I could swear it was accelerating in curves on 6.1.

 

[Saghost]

You also seem to misunderstand the laws of physics a bit. Under almost no conditions would a following distance of "one inch" be safe, even for the fastest computer. Not even at a of couple feet, most likely. The Model S weighs over two tons. If the car in front has a better stopping time then you're going to rear-end them. The car has no way to know the braking power of the car ahead, so the gap between vehicles has to be far enough so that if the car in front hits an invisible brick wall and comes to a dead stop almost instantly you still have enough room to stop safely. The car can calculate this safe distance based on speed and other factors pretty easily.

Hmm...

Assumptions:

The Tesla can stop as quickly as the car in front.

(Not 100% true - the Model S can manage 60-0 in 108 feet in testing here - one of the better sedan results but not the fastest; Motortrend came up with 22 high performance sports cars that can do less than one hundred feet, the best at 93 feet here. More important, this assumption guarantees that if the car in front is abruptly decelerated by hitting something, the Tesla will plow into the back of it.)

Tesla is using the Delphi ESR or a comparable unit, which measures the speed of all objects within the field of view every 50 ms.

(No idea, but the Delphi looks similar to the Tesla pictures I've seen and seems to be the best of the current supplier batch.)

The car can make a decision in 20 ms.

(Pulled from thin air, but I think conservative.)

Bosch iBooster in the Tesla application can develop full braking force in 120 ms.

(vendor certified for the general case.)

The car in front starts braking instantly immediately after a radar ping.

(worst case, almost certainly unrealistic - I don't think any car can switch instantly to full braking without a buildup period.)

This means that the minimum distance where the Tesla can match the braking of the worst case car in front is the distance it covers in 190 ms. At 60 mph (88 feet per second,) that would be (.19*88) 16.72 feet.

Actually, that's quite a bit further than I was thinking it would be before I started pulling numbers together. There is obviously a big hunk of conservatism in the assumption that the other car can slam on the brakes that fast, but it would still be dangerous for Tesla to program anywhere near that close because of the discussion in the first assumption about running over cars that hit things.

Even with the iBooster (3x faster than past systems, they say,) ~2/3rds of this delay is applying the brakes, which implies that a train of communicating cars can benefit substantially over a train of cars that don't talk and just rely on their own sensors (because the communicating cars can start their brake delay before they see the car in front braking.)

Not advocating anything in particular wrt TACC here, just saw a couple folks throwing swags at an interesting technical question.
Walter

 

[green1]

The Tesla can stop as quickly as the car in front.

(Not 100% true - the Model S can manage 60-0 in 108 feet in testing here - one of the better sedan results but not the fastest; Motortrend came up with 22 high performance sports cars that can do less than one hundred feet, the best at 93 feet here. More important, this assumption guarantees that if the car in front is abruptly decelerated by hitting something, the Tesla will plow into the back of it.)

And right there is where the problem lies. You are guaranteeing a collision under certain circumstances. Why would you want that built in guarantee? I may trust my driving, and I may trust the Tesla's driving, but I most certainly do NOT trust the driving of the guy in front of me, especially when the one controlling the brakes is a computer that can't see past the car in front of it.

 

[Saghost]

And right there is where the problem lies. You are guaranteeing a collision under certain circumstances. Why would you want that built in guarantee? I may trust my driving, and I may trust the Tesla's driving, but I most certainly do NOT trust the driving of the guy in front of me, especially when the one controlling the brakes is a computer that can't see past the car in front of it.

As I said in my wrap-up sentence, I'm not advocating changing TACC to follow that closely, just trying to answer the question a couple folks were WAGing at up thread.

 

[green1]

As much as I think that human drivers should also back off further, the computer has even more reason to do so, sure it can act more quickly than a human once it detects a problem, but it can only "see" the car immediately in front, it can't see traffic patterns further ahead, so it's ONLY buffer is the bit between it and the car in front, whereas a normal driver can use other buffers that it can see further ahead.

That said, this thread explains perfectly why we hear about 30 or 50 or even 100+ car pileups, once one car hits something, many of the people in this thread are bound to be in that same collision (note I didn't say accident, it's not accidental if you could have avoided it by following normal safe driving procedures)

 

[billarnett]

I suspect most of those 50 car pile ups are in Limited visibility or limited traction situations.

It might be interesting to see how fast Tesla can ramp up the regen braking. It might be considerably less than the hydrolyic brakes' 120ms. Of course, at present it isn't anywhere nearly as forceful as the friction brakes but it does slow the car some reducing slightly the minimum distance that saghost calculated. And I suspect that the full power of the motor(s) could be applied in an emergency.

 

[green1]

I suspect most of those 50 car pile ups are in Limited visibility or limited traction situations.

Auto pilot *IS* a limited visibility situation. the car can not see more than one car ahead, and as a first responder, let me tell you that many multi-car pileups are not due to either of those situations, but simply due to following too close. On a busy major thoroughfare it is actually quite rare to manage a simple two vehicle collision, there's almost always at least 3 or 4.