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

[ev-now]

Somewhat on-topic, while driving my AP1 car to work today under manual control, I saw a flatbed truck stopped on the shoulder of a very twisty road exactly where the road curves to the left. It was a little surprising to hear/see the collision alarm sound, but I guess it made sense...if I didn't turn to follow the road, my path would indeed have intersected the middle of the truck

Very on topic, as this is exactly the kind of false positive you'd not want it to emergency brake for ... one of those rather too common corner-cases. Just as well you were only looking for assistance from your cars sensors at the time, not for it to make a decision on it's own.

 

[oktane]

Elon Musk had suggested on 12/31/16 that the rest of us clowns would get the limited AP2 firmware this week. Given the calibration issues do you think it's still likely?

 

[MarcusMaximus]

Elon Musk had suggested on 12/31/16 that the rest of us clowns would get the limited AP2 firmware this week. Given the calibration issues do you think it's still likely?

I'm guessing probably not. My impression was that it was this week assuming there weren't any major issues. Losing not just the new features but even normal cruise control permanently is probably a major enough issue to hold off until a fix is issued.

At least I'd hope so.

 

[MP3Mike]

Elon Musk had suggested on 12/31/16 that the rest of us clowns would get the limited AP2 firmware this week. Given the calibration issues do you think it's still likely?

He said: "HW2 Autopilot software uploading to 1000 cars this eve. Will then hold to verify no field issues and upload to rest of fleet next week."

Since there are field issues, I don't expect it to go out fleet wide this week. They have to come up with a new version that fixes the known blocking issue(s), and then probably select an additional 1000 cars to release it to and see if any additional field issues come up before releasing it widely.

I think it will be interesting to see if the extra week(s) give them enough time to enable any other features or improve the existing ones noticeably.

 

[sinv]

In my opinion the expectations are way too high. It is just not possible to gather sufficient safety from so many sensors. Typically such high safety targets can only be reached by extremely robust sensors, and typically you never should use more than 3 of them (otherwise impossible to certify and validate such systems because the possibilities cannot be overlooked). I don't know of any industrial examples which could reach such requirements by cameras!
Just have a look how planes do their autoland or how they measure the true air speed, each of such a malfunction would have catastrophic consequences. The necessary signals all come via one extremely stable physical principle (pressure for air speed, radio waves for height and position). In 2009 even an Airbus from AF went down (flight 447) due to instable sensor signals, however nobody would think about using a (really cheap) gps signal for additional plausibility checks.
Sorry to say this, autonomy will never be reached with cameras and even multiple lidars I am not sure (why else changed google their targets during the last weeks und gave up level 5 at first).

So my thesis is: unnecessary/too much sensors are clearly a disadvantage and cannot be handled appropriate (true only for safety technology of course).

 

[MXWing]

I've been trying to drive as much as I can to provide 'data' but the 35mph autosteer limit and ONLY on highways made it completely untestable to me.

 

[oktane]

I've been trying to drive as much as I can to provide 'data' but the 35mph autosteer limit and ONLY on highways made it completely untestable to me.

Might I suggest a nice trip down the 405 approximately 5:30pm today? j/k

I think you're right they will not release firmware to us this week. I was sad initially that I didn't receive it, but since it only works up to 35MPH anyway you're right the utility is limited. Good for painfully heavy traffic though?

 

[MXWing]

Might I suggest a nice trip down the 405 approximately 5:30pm today? j/k

I think you're right they will not release firmware to us this week. I was sad initially that I didn't receive it, but since it only works up to 35MPH anyway you're right the utility is limited. Good for painfully heavy traffic though?

We'll see how I-5 and 55 goes this afternoon, especially with the rain and all.

TACC has been amazing for stop and go. That will reduce your driving fatigue. With a following distance of 7, control of the wheel and TACC I feel pretty good about not hurting myself or others.

Personally, if I were Tesla I would shelve autosteer until it can do at least 65MPH so people can play and validate it on the far right lanes Instead, spend the effort on the idle object detection for FCW/TACC.

 

[DJ 240V]

I've been trying to drive as much as I can to provide 'data' but the 35mph autosteer limit and ONLY on highways made it completely untestable to me.

Same here . Last eve on 405 i drove in TACC for about 10 mins under 35 miles an hour in peak traffic and no sign of auto steer. This morning it briefly flashed up - but then traffic cleared before I could engage. Where is the traffic when you need it!

 

[RDoc]

In the current revision TACC is only designed to track MOVING cars. If there is standstill traffic in front of you then you'll have to brake manually.

Imagine you're driving a device that can only see things that move. If it hasn't seen it move it's 100% blind to it.

I'm very doubtful that's true.
Do you have a source for that statement?

 

[MXWing]

I'm very doubtful that's true.
Do you have a source for that statement?

Just from observation this is what everyone has been reporting. If the SAME object was moving, and then stopped moving, it will track it.

If you are coming across an immobile object that as not been previously tracked - you will hit it at full speed.

 

[RDoc]

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.

This is explicitly not true. AP1 has no problem identifying stopped objects and, in fact, slowing down and coming to a full stop to avoid hitting them. I've personally experienced this many times in my car.

 

[RDoc]

Just from observation this is what everyone has been reporting. If the SAME object was moving, and then stopped moving, it will track it.

If you are coming across an immobile object that as not been previously tracked - you will hit it at full speed.

That's completely at odds to my personal experience. On several occasions I've had circumstances like coming over a hill with a line of stopped traffic in front of me an the car very nicely slowed down and stopped.

I've heard this on the forums before but from my experience this is an internet fable.

 

[chillaban]

This is explicitly not true. AP1 has no problem identifying stopped objects and, in fact, slowing down and coming to a full stop to avoid hitting them. I've personally experienced this many times in my car.

Even in later 7.1 builds, I found that AutoPilot can identify the vast majority of stationary vehicles. It is true that earlier generations of Autopilot software as well as the ACC software in a lot of vehicles ignore or unreliably detect stationary vehicles, though, and out of an abundance of caution you should not rely on stationary vehicle detection without remaining vigilant.

I've had detected stationary vehicles suddenly disappear off the screen before.

 

[RDoc]

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.

No, it will stop. This whole assertion about the radar not seeing motionless cars is not true.

Think about it for a second, how would the radar even know it was stopped? The radar is moving so unless the target is moving at exactly the same speed, everything it sees is moving.

 

[chillaban]

No, it will stop. This whole assertion about the radar not seeing motionless cars is not true.

Think about it for a second, how would the radar even know it was stopped? The radar is moving so unless the target is moving at exactly the same speed, everything it sees is moving.

There absolutely is a technical reason for the motionless cars assertion, and it is due to the operating principle of doppler radar, which is the primary measuring mode of millimeter wave automotive radar...

If you've ever used a doppler radar gun (like a police radar) before, you'd know that while in motion, the strongest return signature is for your current vehicle speed. Every stopped object that the radar beam bounces off generates a return signature of your vehicle speed, including the parts of the beam that hit the ground, the curb, overhead signs, etc. Bonus trivia: Moving police radar treats the strongest radar signature as the speed of your car, and the next strongest signal as the target vehicle being clocked. There are a lot of patents, especially from Kustom Signals, relating to estimating patrol vehicle speed based off radar signatures in moving mode.

It's a difficult problem to filter out these signatures to find what's a car in your path and what's a harmless object that you'll pass by. And usually the way you do that is by combining radar with a camera system like MobilEye's which provides rough distance estimates based off visual estimation. If the camera says "ok I see a stopped car about 150ft away", you can start correlating that with radar signatures for stationary objects (traveling at negative your speed) that are roughly at that distance.

When a object moves, it's a lot easier of a problem to solve. If you see a fairly large radar return with a speed other than the opposite of your speed, and changing even slightly, you can be pretty sure it's a car, even without any visual confirmation.


The worst corner cases are with weirdly shaped stationary vehicles, such as painted catering vans, garbage trucks, etc. Regardless, the manual still says:

Warning: Tra c-Aware Cruise Control can not detect all objects and may not brake/ decelerate for stationary vehicles, especially in situations when you are driving over 50 mph (80 km/h) and a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you instead.

 

[3Victoria]

Where is the traffic when you need it!

Hey, they've invented a magic traffic-eliminator, sweet!

 

[leletyM3]

Can't wait for the update on mine

 

[MXWing]

That's completely at odds to my personal experience. On several occasions I've had circumstances like coming over a hill with a line of stopped traffic in front of me an the car very nicely slowed down and stopped.

I've heard this on the forums before but from my experience this is an internet fable.

If you won't take my personal experiences for it, how about Tesla's? It either works 100% of the time or it doesn't work at all (BRAKING FOR IMMOBILE OBJECTS). I don't want your Tesla maybe/maybe not rear ending me because you think I'm making up Internet tales.

--
My email and response from Tesla

My Email:

Dear Tesla,

I wanted to write my feedback to you after two days of driving with firmware 2.50.185 with the hopes the information will contribute to the success of the Tesla vision.

False Positive Detection on Forward Collision Detect and Sudden Breaking

1.) Headed towards San Diego super charger from the Midway museum on Sunday Jan 1st.
2.) Forward Collision Detect set to early.
3.) TACC enabled at 70mph.
4.) On 3 occasions I hear beeping and see a red car show up on my dashboard. Car deaccelerstes sharply each time. I hit brake the moment I can react. Nothing is in front of me.
5.) I turn off forward collision assist for the remainder of the trip home and try to evaluate TACC performance independent of collision assist. I never saw the issue again with using TACC only.

My hypothesis is there might of been an overpass or cars in front but "above" me which set something off. I know GPS systems can be confused by objects/road on the Z-axis of travel.

TACC and Immobile Cars in front of it.

I turned on TACC on a road that has a 50mph speed limit and it has lots of red lights.

As TACC was coming up to a stopped car it did not do any slowing down.
I was forced to brake hard. As a human driver I would have seen the car far enough in advance to let off the "gas", And then come to a full stop with very light braking. I do not believe TACC had any reaction to the immobile car.

However, when the light turned green, and we both hit 25 - TACC behaved how I would expect. It maintained sufficient following distance, accelerated and decelerated as needed and could handle full stops and starts from full stops.

Tesla's Response:

Thank you for reaching out with your feedback. Please let us know if your vehicle continues to detect objects which do not appear to be real and we would be happy to have service double check your sensors.

As for the example that you mentioned where you came up upon a stopped vehicle, TACC is not optimized for detecting them at this point. I have pasted the description from the manual which covers this in particular but I do recommend reading the correct operation of the driver assist functions in the manual as well.

"Warning: Traffic-Aware Cruise Control cannot detect all objects and may not brake/decelerate for stationary vehicles, especially in situations when you are driving over 50 mph (80 km/h) and a vehicle you are following moves out of your driving path and a stationary vehicle or object is in front of you instead. Always pay attention to the road ahead and stay prepared to take immediate corrective action. Depending on Traffic-Aware Cruise Control to avoid a collision can result in serious injury or death. In addition, Traffic-Aware Cruise Control may react to vehicles or objects that either do not exist or are not in the lane of travel, causing Model X to slow down unnecessarily or inappropriately."

 

[RDoc]

There absolutely is a technical reason for the motionless cars assertion, and it is due to the operating principle of doppler radar, which is the primary measuring mode of millimeter wave automotive radar...

If you've ever used a doppler radar gun (like a police radar) before, you'd know that while in motion, the strongest return signature is for your current vehicle speed. Every stopped object that the radar beam bounces off generates a return signature of your vehicle speed, including the parts of the beam that hit the ground, the curb, overhead signs, etc.<snip>

To start with, automotive radar capabilities aren't based on the frequency shift that simple doppler radar uses. It does measure frequency shift, but the main way it gets information is by looking at the delay in features of the "chirp" it emits.

I'm extremely doubtful that the strongest echo on any radar system is related to the relative velocity of the target. Whether the object the beam strikes, either directly on echo, is moving or not has nothing to do with the strength of the return, although it would affect the frequency of the return. The reflective characteristics of the surface and its angle to the beam are the main determinants of the return strength.

If you shine a flashlight at a car, does the reflected light intensity change if the car is moving or not?