EAP HW2 - my experience thus far... [DJ Harry]

[brkaus]

I can't find the braking distance for a Genesis at 25 mph, but it is rated as exceptional braking. If I extrapolate from the 60MPH braking distance compared to a standard braking table... I would *guess* it could stop in 20ft or better from 25 mph.

I think the ultrasonic is too slow. What's the beam spread on the radar?

Would it have confidence at 25ft that the object is in it's direct path?

Heck, how directional is the radar for the matter? What kind of image does it give back to the computer?

 

[green1]

I can't find the braking distance for a Genesis at 25 mph, but it is rated as exceptional braking. If I extrapolate from the 60MPH braking distance compared to a standard braking table... I would *guess* it could stop in 20ft or better from 25 mph.

I think the ultrasonic is too slow. What's the beam spread on the radar?

Would it have confidence at 25ft that the object is in it's direct path?

Heck, how directional is the radar for the matter? What kind of image does it give back to the computer?

Radar is pretty low resolution, which is why so many manufacturers insist Lidar is needed for autonomous vehicles. The radar can't tell you the shape of anything, only the rough size and position as well as the speed. The trick is that radar can tell where the object is, but not where the road will go. So it doesn't know the difference between a car in front of you in a straight line, or a car parked on the outside of a curve that is no issue. That's where the camera comes in, it knows where the road is going. Lidar helps a small amount by giving a more accurate shape to the object, making it easier to recognize, and easier to get close to without hitting.
Personally I think Radar + Camera should be enough, but you can't use one or the other, you need to use both (and that includes in directions other than the front)

 

[brkaus]

Radar is pretty low resolution, which is why so many manufacturers insist Lidar is needed for autonomous vehicles. The radar can't tell you the shape of anything, only the rough size and position as well as the speed. The trick is that radar can tell where the object is, but not where the road will go. So it doesn't know the difference between a car in front of you in a straight line, or a car parked on the outside of a curve that is no issue. That's where the camera comes in, it knows where the road is going. Lidar helps a small amount by giving a more accurate shape to the object, making it easier to recognize, and easier to get close to without hitting.
Personally I think Radar + Camera should be enough, but you can't use one or the other, you need to use both (and that includes in directions other than the front)

Good input on radar, thx. I agree 100% with you on camera +. But at 20-25ft a lot of assumptions could be made. Wondering (foolishly perhaps) if that's how some of the other systems are successful at low speeds.

 

[oktane]

I've had adaptive (radar) cruise control on my cars for years. Yes sometimes bridges etc. create a collision warning or will slow the car, but I've never had the car slam on the brakes for this. It happens around turns mostly.

Tesla's logic should be to treat all stationary echoes as a threat, until proven otherwise by the camera system. I'll take a nuisance slowdown over being blind to a stopped car around a turn. Hopefully the camera image recognition algorithms will improve this function. Sad that someone has already had to die because of this oversight or technological limitation (not sure which).

 

[DJ 240V]

What distance setting were you on?

(the end of the TACC stalk rotates to increase/decrease distance. On AP1 cars you have options of 1-7, probably the same on AP2.)

At one point, I was arguing for a acceleration profile setting like you're describing; after some reflection, I think the distance setting is enough:

Presumably, anyone who sets the car to distance one is looking for an aggressive acceleration/deceleration profile - as close as the car can safely manage to holding the exact minimum safe distance behind the car in front no matter how that car moves.

Similarly, someone who sets the car to 7 is clearly looking for a smooth, relaxed experience with minimal connection to the cars around them.

I haven't seen clear evidence of it happening, but what I'd like to see is Tesla adjusting the car's acceleration and braking behavior based on the distance setting (along with changing the target distance, of course.) the closer you set, the quicker it becomes in reacting to the environment and the less smooth it is (which is handy, because when it comes to braking that's pretty much what's required for safety anyway.)

Thats a good point . I started at 7 and then quickly settled b/w a 3 or 4. As 7 on a highway will end up people constantly cutting you off - but let try that and see if that makes it any smoother.

 

[MarcusMaximus]

Sad that someone has already had to die because of this oversight or technological limitation (not sure which).

They didn't. They died because they were watching a movie while driving and assumed the driver assist feature would save them from any possible situation. Normally that may even be a safe bet, but that's still the risk you take when doing such things until we get true autonomous cars.

 

[somnambule]

This video seems to suggest that 8.xx can detect a stationary person. Test #1 is to have a person stand in the middle of the road and have the car on AP go at 20mph.

Unless the person was already within radar range when he walked on to the road, this seems to suggest that AP1 with newer software is able to stop for at least certain types of stationary objects (the text in the video says 7.xx did not stop in this case).

It's unclear if the same behavior would be observed under AEB when not in AP.

 

[spiderbro]

They didn't. They died because they were watching a movie while driving and assumed the driver assist feature would save them from any possible situation. Normally that may even be a safe bet, but that's still the risk you take when doing such things until we get true autonomous cars.

But the Tesla salesperson didn't tell me I couldn't do that! - says every whiner in the Breaking: HW2 thread.

 

[Saghost]

This video seems to suggest that 8.xx can detect a stationary person. Test #1 is to have a person stand in the middle of the road and have the car on AP go at 20mph.

Unless the person was already within radar range when he walked on to the road, this seems to suggest that AP1 with newer software is able to stop for at least certain types of stationary objects (the text in the video says 7.xx did not stop in this case).

It's unclear if the same behavior would be observed under AEB when not in AP.

FW 8.0 does an excellent job of stopping for stationary objects 99%+ of the time on AP1 cars, including stopped cars that it never saw move.

I've only had one occasion when it didn't - or more accurately, when it hadn't started braking at a point that I got really worried and overrode it. It might still have stopped that time, too, if I'd waited, but I couldn't be sure. (traffic light with stopped cars that had been masked by geometry until well after they stopped.)

The problem is that some folks assume it'll do that 100% of the time, and it can't until the whitelist becomes active and you're on a whitelist route. Autopilot is far superior to any other driver assistance system I'm familiar with, and continues to improve.

It can't handle everything, though, and folks who assume it will tend to end up in an accident sooner or later - and Tesla and the rest of us learn more about that corner case as a result and they find ways to improve the system.

 

[Economite]

It can't handle everything, though, and folks who assume it will tend to end up in an accident sooner or later - and Tesla and the rest of us learn more about that corner case as a result and they find ways to improve the system.

That seems like a really dangerous way to "discover" corner cases; especially since most of the situations discovered in this manner will be things that have already been discovered by Tesla (and perhaps discussed in these forums) but just not publicized to owners in a way that they will understand and seek to avoid.

 

[theslimshadyist]

Ok, hopefully he will read and reply!

 

[drklain]

Radar is pretty low resolution, which is why so many manufacturers insist Lidar is needed for autonomous vehicles. The radar can't tell you the shape of anything, only the rough size and position as well as the speed. The trick is that radar can tell where the object is, but not where the road will go. So it doesn't know the difference between a car in front of you in a straight line, or a car parked on the outside of a curve that is no issue.

Is this based on your knowledge of radar in general or the radars used by cars or the radar used by Tesla in specific? The reason I ask is that, from a radar perspective in general, none of what you are saying is true.

• The resolution of radar is a function of its frequency (which is related to wavelength). From a physics perspective, you can't have resolution smaller than the wavelength because, well, physics. The problem is, shorter wavelengths give higher resolution but can detect at shorter ranges...because, well, physics. Millimeter wavelength radar can be VERY, very high resolution down to millimeters and are absolutely capable of telling you the shape of something. Those systems aren't science fiction -- they exist today and are used for a variety of applications. You know those TSA body scanners everyone was so upset about because they showed such an accurate depiction of what people looked like so as to be a "nude photo"? Guess what -- that's a millimeter wave radar!
  
• Radar in and of itself cannot tell you where an object is going but through use of radar return mapping, a very precise directional vector and speed can be calculated in three dimensions. It can then predict where the object is going to be in the future (again in three dimensions of the system is sophisticated enough). If an object is turning, a system is also capable of calculating the rate of turn and then calculating the curve of the turn and what the predicted path of the object is. Of course a change the objects behavior can change that curve, but some radars update hundreds of times per second, so the data is very accurate. What the radar system CAN'T predict is operator intent or future control inputs.
• Yes, with enough programming, a radar processing system could tell the difference between a car in front of you in a straight line or a car parked on the outside of a curve that is no issue -- it all depends on how you are defining "radar system". Are you defining it as simply the transmit/receiver and the antenna? Or do you include the signal processing system that is driving that?
• There are white papers published talking about millimeter-wave radars being used in automotive applications which make sense (they don't need to detect stuff 200+ miles away, they are looking at distances of inches, feet and maybe a mile or two.

For examples (and the only stuff you can find on the internet is the unclassified stuff) google synthetic aperture radar, inverse synthetic aperture radar, millimeter-wave radar, W Band Radar and FM-CW radar.

@green1 -- You obviously feel strongly about Tesla and are pissed off about how the AP software rollout has taken place. I get it and agree that Tesla has not handled things with AP and EAP and FSD in a way I personally would have done so or liked. In these forums you keep making assertions that don't seem to match physics or logic.

My background is 28 years in the military and tech industry including operating and association with some of the most sophisticated automated weapons sensor and control systems on the planet as well as DARPA and US National Laboratory programs which were on the cutting edge of research and technology as Chief, Concepts and Experimentation of a Defense Agency that funded most of those programs in the United States and post-government employment in a tech company that was doing state-of-the-art cyber work. That said, I am 100% certain that several people on this forum are far more experienced and knowledgeable than I am (Bruce (@bmah) is one of them). Might I ask -- Do you have any background in software design, artificial intelligence, automation, sensors (optical, ultrasonic or radar) or systems engineering or are all the facts you are spouting simply your opinions based on no first hand knowledge or training?

 

[DJ 240V]

Update : got this clarification from Tesla

As for Automatic Emergency Braking this is only active while TACC is active at this time. This will change as software updates are released but just as the other features are limited this will be for now as well.


I assumed this was available all the time , i.e. TACC or not , however got this clarification from Tesla - so stay safe and don't assume you have this yet!

 

[DJ 240V]

Hi Harry,

I got the firmware as well and wanted to comment on your experiences.

1.) I honestly think TACC speed settings would add just one more layer of complexity. Cars on the side just need to be aware that Teslas are really damn fast and they should plan on changing lanes behind you. As far as how TACC will behave:

a.) Large gap with no object in front. Max acceleration, and slow down once a moving object is in front.
b.) Small gap (assuming small gap means object in front). It will accelerate either slowly or quickly depending on if the gap gets smaller or bigger. Once TACC locks on, it will adjust speed accordingly to match the distance profile (1-7). It wont gun it if an object is close by.

2.) Rather than hover your foot on the brake, plant it on the ground in front of the brake pedal. You will release relax more, release tension on your body, reduce fatigue and most importantly, free up your posture so you can be more like a captain. Scanning around for the big picture.

If needed, just raise your foot and bring it down on the brake.

Good tip - thanks. One thing for sure I'm not driving subconsciously anymore !
Im trying to do #2 but do not trust the system yet or will not for a long time. It's often stops a whole second or two before I would have eased off the accelerator . I also feel like it uses the brakes way more than I would ( single foot driving is smooth and fun!)

 

[DJ 240V]

Just remember it was also a few months away in October 2014 for AP1... Those owners are still waiting too.

I did mean "Elon Months"

 

[DJ 240V]

I'm a little confused by the failure to identify stopped cars. I know that is how AP1 works, but it's also one of its most dangerous issues.

I thought the entire point of "enhanced" AP was that they had solved this issue. Remember the whole "we don't need lidar" thing? Lidar tells you where there is a stationary object and when it's in your path vs off to the side. But the claim was they had determined they could do the same with cameras and didn't need it.

Perhaps it's on the way, but it's a little confusing.

So just to clarify - If AP1 used TACC on a freeway that had no one in front and a car was stopped up ahead - it would not stop for it when it was in range ?

On EAP I haven't experienced the scenario yet as the range is supposedly doubled and I have only tried it with traffic.

 

[ev-now]

What distance setting were you on? ....
I haven't seen clear evidence of it happening, but what I'd like to see is Tesla adjusting the car's acceleration and braking behavior based on the distance setting (along with changing the target distance, of course.) the closer you set, the quicker it becomes in reacting to the environment and the less smooth it is (which is handy, because when it comes to braking that's pretty much what's required for safety anyway.)

You've seen no evidence, because it does not work that way when it loses it's locked car in motion. At rest it pulls away apparently slower when set at 7, as it waits for the gap before accelerating. When the car in front 'disappears' it still puts it's (!) foot down hard until it reaches whatever speed is set. I've compared 7, 5 and 4 - not happy at less than 4 - so it might get even more lead footed at 3 or less, but 7-4 behaves the same for me (AP1).

 

[ev-now]

The problem is that some folks assume ...

That these things are more than Driver Assistance - and that about sums it up. Nobody with AP1 or EAP will get all corner-cases covered, and it think FSD will be way longer out than people hope and want.

 

[ev-now]

That seems like a really dangerous way to "discover" corner cases; especially since most of the situations discovered in this manner will be things that have already been discovered by Tesla (and perhaps discussed in these forums) but just not publicized to owners in a way that they will understand and seek to avoid.

Not really, if they read the manual and follow the MANY warnings about the limitations of the systems in use - which in many cases far exceed the competition and a few might be a little behind. Those claiming other cars do way better should read the manuals of those vehicles to see very similar warnings as their manufacturers also do not believe they've covered all corner-cases!

 

[ev-now]

On a positive note (as far as I am concerned). I was dring my 2015/AP1 today in blowing snow, in the dark with all driving assistance switched off (no AP/TACC) and following the car in front at what I believe to be a safe distance (as did my passenger) when the system squealled and flashed a red car ahead of the one in front and applied the brakes. Car in front moved right, and revealed the stationery car in front.

Yes, this was a slow motion version of the recent video from the Netherlands, but in far worse conditions.

Had the vehicle in front NOT been able to stop, this might have caught me out as my distance was obviously not based on the assumption of a sudden, complete stop of the vehicle in front. Didn't save me from anything, but reassured me that AEB, even in poor conditions and with AP1 hardware can do a great job of spotting an obscured stationery vehicle.