Autonomous Car Progress

[Tam]

When will Tesla RADAR be able to reliably react to dangerous stationary obstacles appropriately?

The article below seems to suggest unreliability is RADAR's nature so Tesla needs to add LIDAR to its suite:

Why Tesla's Autopilot Can't See a Stopped Firetruck

 

[boonedocks]

Waymo’s self-driving minivans are coming to Atlanta
Waymo’s self-driving minivans are coming to Atlanta

Georgia has a pretty open law for FSD vehicles. Hopefully when Tesla begins limited testing of FSD we will be one of those states able to take advantage of it. Regulatory approval is NOT Tesla's hurdle for Georgia...the hurdle will be Tesla

 

[d21mike]

When will Tesla RADAR be able to reliably react to dangerous stationary obstacles appropriately?

The article below seems to suggest unreliability is RADAR's nature so Tesla needs to add LIDAR to its suite:

Why Tesla's Autopilot Can't See a Stopped Firetruck

I do not know but really hope soon.

But the article also says that Elon does not agree that LIDAR is required.

 

[Tam]

But the article also says that Elon does not agree that LIDAR is required.

After Florida Autopilot fatal accident, Elon Musk believed that Tesla could tweak current RADAR "to create a coarse point, like LIDAR."

Four months after that, Tesla blog reported that it did it and how its LIDAR-like RADAR works:

"The second part consists of assembling those radar snapshots, which take place every tenth of a second, into a 3D "picture" of the world. It is hard to tell from a single frame whether an object is moving or stationary or to distinguish spurious reflections. By comparing several contiguous frames against vehicle velocity and expected path, the car can tell if something is real and assess the probability of collision."

At that time, when asked if this could prevent the Florida Autopilot accident, Elon Musk replied:

"These things cannot be said with absolute certainty, but yes, we believe it would have."

It sounds like Tesla has claimed that it has solved the unreliability of RADAR problem by making it LIDAR-like without buying a real LIDAR, so of course, you are correct that Elon would not agree that LIDAR is required.

 

[scottf200]

It sounds like Tesla has claimed that it has solved the unreliability of RADAR problem by making it LIDAR-like without buying a real LIDAR, so of course, you are correct that Elon would not agree that LIDAR is required.

I wonder if the HW2.5 "Continental ("Radar Front Conti") with "primary" and "secondary" CAN-signalling to AP2.5ECU" is any better at this processing.

It looks like it handles pedestrians better as an example of functionality: HW2.5 capabilities

 

[lunitiks]

Taking this from memory, but according to conti’s spec sheets it seems like the radar has two «FOVs», one short and one long range, with different attributes. Don’t remember where, but I speculated that the primary and secondary could be related to this somehow, perhaps it’s rigged so that both ‘beams’ can be used and processed simultaneously.

I do not think the ‘secondary’ canbus is purely for redundancy. Wiring diagrams use the word ‘redunant’ in other aspects but not here. Anyway it would be nice to hear more about this from anyone with more automotive radar knowledge

 

[NerdUno]

After Florida Autopilot fatal accident, Elon Musk believed that Tesla could tweak current RADAR "to create a coarse point, like LIDAR."

Four months after that, Tesla blog reported that it did it and how its LIDAR-like RADAR works:

"The second part consists of assembling those radar snapshots, which take place every tenth of a second, into a 3D "picture" of the world. It is hard to tell from a single frame whether an object is moving or stationary or to distinguish spurious reflections. By comparing several contiguous frames against vehicle velocity and expected path, the car can tell if something is real and assess the probability of collision."

At that time, when asked if this could prevent the Florida Autopilot accident, Elon Musk replied:

"These things cannot be said with absolute certainty, but yes, we believe it would have."

It sounds like Tesla has claimed that it has solved the unreliability of RADAR problem by making it LIDAR-like without buying a real LIDAR, so of course, you are correct that Elon would not agree that LIDAR is required.

No mention of avoiding stopped fire trucks so I guess Tesla's design is good to go. Oh wait...

Tesla Autopilot is under federal scrutiny again after Model S crash

 

[Bladerskb]

After Florida Autopilot fatal accident, Elon Musk believed that Tesla could tweak current RADAR "to create a coarse point, like LIDAR."

Four months after that, Tesla blog reported that it did it and how its LIDAR-like RADAR works:

"The second part consists of assembling those radar snapshots, which take place every tenth of a second, into a 3D "picture" of the world. It is hard to tell from a single frame whether an object is moving or stationary or to distinguish spurious reflections. By comparing several contiguous frames against vehicle velocity and expected path, the car can tell if something is real and assess the probability of collision."

At that time, when asked if this could prevent the Florida Autopilot accident, Elon Musk replied:

"These things cannot be said with absolute certainty, but yes, we believe it would have."

It sounds like Tesla has claimed that it has solved the unreliability of RADAR problem by making it LIDAR-like without buying a real LIDAR, so of course, you are correct that Elon would not agree that LIDAR is required.

isn't it intriguing that the likes of @stopcrazypp who vowed that elon will have L4 in 2018 haven't shown their faces around here for months.
but just a few months ago were parading around this very forum proclaiming that Elon will have L4 finished this year and how Tesla are years ahead of the competition in SDC. But now that the jig is up, they are a missing person case.

The current state of AP today shows you exactly what elon tweet were, just another fairy tale. all that coarse radar for better lidar nonsense. I tried to warn you people but you won't listen.

 

[S4WRXTTCS]

isn't it intriguing that the likes of @stopcrazypp who vowed that elon will have L4 in 2018 haven't shown their faces around here for months.
but just a few months ago were parading around this very forum proclaiming that Elon will have L4 finished this year and how Tesla are years ahead of the competition in SDC. But now that the jig is up, they are a missing person case.

The current state of AP today shows you exactly what elon tweet were, just another fairy tale. all that coarse radar for better lidar nonsense. I tried to warn you people but you won't listen.

You're still on about that?

You're certainly not the only one who thought FSD was a farce. I considered it to be one of the two worst things to have happened in 2016. I know I'm not alone in this. Now you're more of Tesla skeptic than I am; I'll grant you that. So maybe there is something in you that compels you to continue on.

I tend to be a skeptic of the entire idea of L4 driving. I'll be that way until cruise automation or some other company comes to Seattle where I can actually ride in one on a routine basis.

You, and I might have different opinions/views but we both know why we're here. We're here because Tesla still has the most interesting thing going on. Sure they're massively late on EAP, and the whole FSD was basically vaporware. But, even small things like a rain detecting neural net is better than nothing for us tech obsessed people. Granted it's been pretty boring since the guy feeding us all that awesome info got pissed at his Tesla and had Tesla buy it back. That was a bit embarrassing I must admit. So maybe Tesla should focus a bit more on just making a solid car, and not worry so much about this EAP/FSD stuff.

Getting back to the point cloud radar I really wonder if that entire idea was scrapped. On the Tesla blog they said they would initially roll it out in firmware 8.0, and then over the course of time they'd built a white list. Where things like over head signs, bridges, etc could be white listed. That way the car could stop for a stopped object in the road, and not signs and whatnot.

Then they'd roll it out where it activated on this new data. But, we haven't heard anything official from Tesla since then. So we have no idea if it's activating on it.

So it seems like they tried it, and concluded that it wouldn't work.

It's also interesting that they switched Radar vendors, and so AP2.5 has a different radar than AP1/AP2.

 

[stopcrazypp]

isn't it intriguing that the likes of @stopcrazypp who vowed that elon will have L4 in 2018 haven't shown their faces around here for months.
but just a few months ago were parading around this very forum proclaiming that Elon will have L4 finished this year and how Tesla are years ahead of the competition in SDC. But now that the jig is up, they are a missing person case.

The current state of AP today shows you exactly what elon tweet were, just another fairy tale. all that coarse radar for better lidar nonsense. I tried to warn you people but you won't listen.

Sigh...
I thought you got killed after you failed to find the post, but you still like to keep calling me out for no reason. Reminder of our last interaction and the nth time I warned you:

First of all elon said that level 5 summon and fsd will be ready in less than 2 year.
Keyword: LESS.

He said this around june 2016. Less doesnt mean you add 1 more year to his statement.

Less = Less than June 2018

Now you are trying to deflect and say u didnt vow l5 fsd will be ready no matter the regulation in 2018. Bro u wont duck this.

I will dig up that post even if it kills me. You wont get away this easy.
Ofcourse thats if u havent edited/deleted it.

I'll say it again, either find the post or stop calling me out.

 

[AnxietyRanger]

Granted it's been pretty boring since the guy feeding us all that awesome info got pissed at his Tesla and had Tesla buy it back. That was a bit embarrassing I must admit. So maybe Tesla should focus a bit more on just making a solid car, and not worry so much about this EAP/FSD stuff.

He's back: Rumor summary: Blind-spot cameras, Rain sensing, Level 3, Big battery, Interior/HUD #70

I tend to be a skeptic of the entire idea of L4 driving.

The AP2 HW was actually sold to us Level 5 capable...

Tesla's Full Self Driving level? #13

 

[AnxietyRanger]

@stopcrazypp

Well, at the very least none of these things happened:

So what Elon means is that in 3-6 months there will be a noticeably different feature for those that opted for the FSD vs those that just opted for the EAP. It does not mean that they will be releasing FSD (level 5) driving at this point. EAP uses 4 cameras, FSD uses 8 cameras, so there are plenty of other features they can release for FSD (not related to level 5 driving) that won't be available in EAP.

The only time-line related to level 5 is that there will be a cross country demo by the end of the year. But there is no promised timeline for when level 5 driving will be released to the public.

Except for the last sentence of course.

 

[zmarty]

The 2017 California DMV disengagement reports are online: https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017

For Reporting Year 2017, Tesla did not test any vehicles on public roads in California in autonomous mode, as defined by California law.

This, to me, is beyond bizarre.

 

[scottf200]

The 2017 California DMV disengagement reports are online: https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017

For Reporting Year 2017, Tesla did not test any vehicles on public roads in California in autonomous mode, as defined by California law.

This, to me, is beyond bizarre.

Did you see the GM Cruise report --- lot of reasons described as "Other road user behaving poorly"

 

[EinSV]

Did you see the GM Cruise report --- lot of reasons described as "Other road user behaving poorly"

Accidents every 6K miles? I liked their concept of driving in challenging environments to learn faster but this is not good. Waymo and GM still lead the pack in California’s new self-driving report cards

 

[Bladerskb]

Sigh...
I thought you got killed after you failed to find the post, but you still like to keep calling me out for no reason. Reminder of our last interaction and the nth time I warned you:


I'll say it again, either find the post or stop calling me out.

There were multiple posts where you eluded to tesla providing FSD in 2018 barring regulations.
It doesn't because tesla has failed to even implement something as simple was EAP.
Its quite obvious that i'm right and you are wrong.

Now this CA disengagement simply told us what we already know.

 

[stopcrazypp]

There were multiple posts where you eluded to tesla providing FSD in 2018 barring regulations.
It doesn't because tesla has failed to even implement something as simple was EAP.
Its quite obvious that i'm right and you are wrong.

Now this CA disengagement simply told us what we already know.

It seems you will never stop putting words in my mouth. Again, find the post or stop using me as a strawman to toot your own horn. If there were multiple posts it should be easy to find, much less one post. You had months to find it since our last interaction.

 

[cwerdna]

Regarding Waymo, there's Waymo drove 2 million autonomous miles in 2017.

 

[zmarty]

Not all press demos go smoothly...

We were in an accident during an automated driving tech demo

 

[R.S]

Accidents every 6K miles? I liked their concept of driving in challenging environments to learn faster but this is not good. Waymo and GM still lead the pack in California’s new self-driving report cards

We are at a pretty critical point right now, the car isn't as good as a human driver and doesn't learn nearly as fast, but it's good enough that the person behind the wheel gets distracted too easily.

I guess a perfect human would have stepped in and avoided every one of those accidents, but just imagine spending 6k miles behind the wheel of an AV, in urban conditions. That's a lot of time where the car drives with no issues, so I guess you can't fault the test driver to not be able to react once things don't work out.

And it will get even worse as the car gets better. I guess soon there will be no disengagements and just accidents.

Of course if you have a disengagement every 3 miles, as Tesla had in 2016, there will be mostly disengagements and much fewer accidents, since the driver has to take over regularly.