Why AP 2.0 Won't Be Here Soon, and It Won't Be What You Think It Is

[roblab]

You said new cars roll off the line. I assumed you meant production line. Which mine is off of said line. Meaning my car would not have 2.0 hardware. There has to be a cut off somewhere. I could look at my car. Determine no new hardware and walk away right?

What usually happens is that Tesla announces that they have been doing such and such hardware since "last year". They didn't tell anyone. They just switched. They make 20 improvements a week. Then they release the software and tell everyone that you've had the hardware for the last 6 months.

I guess you could go check your car and hope you are wise enough to see if you have 2.0 hardware. But then there's 3.0. Somewhere, you just have to be happy with what you've got, cause in a year, you won't have it all, again.

 

[Lex]

I am gud at sum things, like summarizing. So like.. may I channel:

We moved away from MobilEye when they became like Mobil.

We have visionary partners !! Or, we can do our own thing.

We need new hardware. But we have the legacy stuff, too.​

 

[number12]

Well, maybe sometimes but they quoted Mobileye noting: “Today we are already preparing with one of the OEM, a first vehicle based on 8 cameras, one radar and ultrasonic around the vehicle."

The implication was that this could be AP 2.0, but (as I argued at the time), it was a Nissan test car that was announced around that time. It literally had 8 cameras running 5 Mobileye EyeQ3.

My point being, I don't believe there was ever any plan for AP 2.0 with 8 cameras and one radar from Tesla by the end of 2015. But yes, I do believe good intentions can slam into reality.

The camera housing was updated right? And there's an extra spot for an extra wire. Looks like they were planning on something

 

[MarkS22]

The camera housing was updated right? And there's an extra spot for an extra wire. Looks like they were planning on something

Well, only the X housing has possible space for one more camera. That's a far cry from a three camera trifocal cluster and five other cameras around the car that was proposed. There have never been any signs of the six other camera locations as noted in Mobileye's statement.

I'm not saying Tesla never had any update plans that may have changed. Just that people have had varying levels of evidence and speculation on AP 2.0 for over a year now. I mean, this article came out less than a month after AP 1.0's Autopilot was turned on.

They're not going to commit to a whole new fleet of sensors without getting real world data back. If they had rushed out the 8-camera, 1-radar AP 2.0, they'd be in a serious bind right now. The longer they wait, while continuing to fleet learn and improve software, the better AP 2.0 will be.

 

[Canuck]

The longer they wait, while continuing to fleet learn and improve software, the better AP 2.0 will be.

I agree with a lot of what you say, but this part lost me since drawn to its logical conclusion (or perhaps illogical one, you tell me) there's no reason to push the boundaries and get things out, until near perfection. That's not the Tesla of the past, and I sure hope it's not the Tesla of the future.

I really think the NHTSA investigation has put any new hardware on hold since why undertake voluntary that which you can negotiate as a settlement? If you do, you have taken that option off the table and thus risk other potential orders.

 

[MarkS22]

I agree with a lot of what you say, but this part lost me since drawn to its logical conclusion (or perhaps illogical one, you tell me) there's no reason to push the boundaries and get things out, until near perfection. That's not the Tesla of the past, and I sure hope it's not the Tesla of the future.

I really think the NHTSA investigation has put any new hardware on hold since why undertake voluntary that which you can negotiate as a settlement? If you do, you have taken that option off the table and thus risk other potential orders.

Yes, I should clarify. While it's true that the longer they wait, the better, you obviously need to launch eventually. That said, they're still the most advanced (as tested by several publications) and they're still making advancements. So there's no immediate need to rush a sensor launch. With 8.0, the majority of the public will assume it is a new sensor suite with magical new radar. Even those that know better will (hopefully) be impressed at the software improvements to Autopilot 8.0 and 8.1.

Back when I originally thought AP 2.0 was premature (around the time of the Model X launch), I didn't think they'd launch a new sensor suite only 2-3 months after launching the original software. I proposed they'd want to gather real world data and maximize the current system before splintering the hardware.

Now we're in a new situation. Before the split with Mobileye, there was talk of that "8-camera, 1-radar, 5 EyeQ3" system potentially being AP 2.0. Mobileye openly talked about its roadmap of a trifocal camera cluster up front so it was assumed Tesla would use that. I believe Tesla was going to do that until they realized even a trifocal camera might not have been able to detect the recent accident. So, they shifted gears, moved away from Mobileye, and worked with Bosch to develop this new radar method.

I believe, given how they have no competition capable of OTA driving assistance updates (to my knowledge?), it's in their best interest to milk the current sensors and test the new radar/whitelisting setup before committing to a new sensor suite. Based on everything we know today, I would guess launching AP 2.0 on the S & X a few months before the 3 is released seems realistic. That seems like a sweet spot between waiting/learning and having a new sensor suite.

The big maybe is that they get enough data to be confident of the new sensors required by December and announce them with Model 3 Part Deux, with immediate availability (but no new functionality yet) on the S and X.

 

[Troy]

On 11 Aug 2016, Electrek published the following:

Sources with knowledge of the Autopilot program told us that the new suite will keep the current front-facing radar and add more around the car, likely one in each corner. Source

A month later, on 11 Sep 2016, Tesla published this article:
Upgrading Autopilot: Seeing the World in Radar

The entire article about the new radar software adds credibility to the rumor a month ago. It sounds like Electrek's anonymous source was correct about multiple radars (potentially 5 radars). Therefore it is likely that the source is also correct about 3 cameras. My best guess is, Tesla has actually built a test vehicle with 5 radars and 3 cameras and this person saw that vehicle. Elon said what they have been working on blows his mind and it will blow other people's mind too. Elon must be driving this test vehicle.

In the conference call on 11 Sep 2016, Elon said the following:

It’s something that I wanted to do for a while. Probably since late last year, but I was always told that it wasn’t possible, you can’t do it, it’s not gonna work, nobody else has made it work, software is too hard, sensor is not good enough, but I really pushed hard on questioning all those assumptions last 3 or 4 months. Like there got to be a way to make this work and now we believe that there is. Source 1 (transcripts) , Source 2 (audio at 32:28s)

It sounds like they built the test vehicle in late 2015 and they have been working on it since then but only managed to make the radars paint a picture in June 2016 after Bosch provided a new firmware that enables access to raw data. What Elon mentions as Tesla vision must be the work they have been doing on this 3 camera setup.

I think initially Elon was not planning to release Autopilot 8.0 firmware update but after the fatal accident he asked the team if they could adapt the new radar software to Autopilot 1.0 hardware as well and they managed to do that because they already had this working for last few months in their Autopilot 2.0 test car.

My prediction hasn't changed. I still think January 2017 is likely for new hardware. February is also a close contender. I just wanted to add some more context why I think so. I can't imagine Tesla postponing Autopilot 2.0 any further than Feb 2017 because it is getting too close to Model 3 preparation time. They will be very busy with all sorts of Model 3 related stuff.

 

[3Victoria]

I think the AP2 hw will be sooner. They need 4x radar since the ultrasonic sensors do not have enough range. 360 video would be nice too. The 3d radar mapping is a quick fix for the cross-traffic problem but also more generally useful. The sw will come later in stages.

 

[MarkS22]

On 11 Aug 2016, Electrek published the following:


A month later, on 11 Sep 2016, Tesla published this article:
Upgrading Autopilot: Seeing the World in Radar

The entire article about the new radar software adds credibility to the rumor a month ago. It sounds like Electrek's anonymous source was correct about multiple radars (potentially 5 radars). Therefore it is likely that the source is also correct about 3 cameras. My best guess is, Tesla has actually built a test vehicle with 5 radars and 3 cameras and this person saw that vehicle. Elon said what they have been working on blows his mind and it will blow other people's mind too. Elon must be driving this test vehicle.

In the conference call on 11 Sep 2016, Elon said the following:


It sounds like they built the test vehicle in late 2015 and they have been working on it since then but only managed to make the radars paint a picture in June 2016 after Bosch provided a new firmware that enables access to raw data. What Elon mentions as Tesla vision must be the work they have been doing on this 3 camera setup.

I think initially Elon was not planning to release Autopilot 8.0 firmware update but after the fatal accident he asked the team if they could adapt the new radar software to Autopilot 1.0 hardware as well and they managed to do that because they already had this working for last few months in their Autopilot 2.0 test car.

My prediction hasn't changed. I still think January 2017 is likely for new hardware. February is also a close contender. I just wanted to add some more context why I think so. I can't imagine Tesla postponing Autopilot 2.0 any further than Feb 2017 because it is getting too close to Model 3 preparation time. They will be very busy with all sorts of Model 3 related stuff.

Only it wasn't a blind rumor. Elon was already talking about radar on Twitter in July: "Working on using existing Tesla radar by itself (decoupled from camera) w temporal smoothing to create a coarse point cloud, like lidar"

This article also claims the trifocal camera housing was already being installed in production vehicles, meaning before August 11th. If the source was correct, shouldn't we have seen this in P100Ds (which were announced August 23rd)?

I'm curious about the approach they'll take. I'd think it would be all or nothing like the initial AP sensors. If they had new camera housings but no new radar, that could seriously hurt sales while people wait for the full sensor suite or upgradability to be confirmed. It's one thing for us to guess. It's another to see evidence of the transition coming.

That said, considering the track record of the site, I do believe there is truth within that article even if the timing isn't exact. Your prediction could align with the second part of the Model 3 reveal.

 

[technomation]

I haven’t seen much in these posts that emphasizes the “competition factor.” It’s hard to imagine that Tesla would hold back on AP 2.0, when the German automakers are coming on so strong and so fast. We’ve seen lots of articles on the rapid advances in autonomous features over the past few months. Lately the opinions are that Mercedes in particular is on par with Tesla, see e.g., today's Car Connection:

“On that front, [the 2017 E-Class] is a magnitude ahead of rivals like the Audi A6, BMW 5-Series, Lexus GS, Jaguar XF, and Cadillac CTS. It's now locked in a headline war with the American-made Tesla Model S for the smartest car on the road.”

Is it really consistent with Tesla’s corporate culture to let themselves be overtaken like this? Seems like they’d be burning the midnight oil to accelerate AP 2.0 development!

http://www.thecarconnection.com/overview/mercedes-benz_e-class_2017#src=10065

 

[MarkS22]

I haven’t seen much in these posts that emphasizes the “competition factor.” It’s hard to imagine that Tesla would hold back on AP 2.0, when the German automakers are coming on so strong and so fast. We’ve seen lots of articles on the rapid advances in autonomous features over the past few months. Lately the opinions are that Mercedes in particular is on par with Tesla, see e.g., today's Car Connection:

“On that front, [the 2017 E-Class] is a magnitude ahead of rivals like the Audi A6, BMW 5-Series, Lexus GS, Jaguar XF, and Cadillac CTS. It's now locked in a headline war with the American-made Tesla Model S for the smartest car on the road.”

Is it really consistent with Tesla’s corporate culture to let themselves be overtaken like this? Seems like they’d be burning the midnight oil to accelerate AP 2.0 development!

http://www.thecarconnection.com/overview/mercedes-benz_e-class_2017#src=10065

This is a software and infrastructure battle. Not commodity sensors. Tesla is burning the midnight oil with software improvements and fleet learning. There's still a lot of work and data to acquire.

I don't think anyone is suggesting they wait around for 2.0 hardware for no reason.

(And, no, Drive Pilot isn't anywhere close to Autopilot.)

 

[Canuck]

This is a software and infrastructure battle. Not commodity sensors. Tesla is burning the midnight oil with software improvements and fleet learning. There's still a lot of work and data to acquire.

Regardless of whether that is true or not, perception often trumps reality, and Mercedes is telling this to the public... I doubt Tesla will sit back and let this continue to be true for very long...

 

[Canuck]

A couple more from the friendly competition...

"Q4 2014 Autopilot Vintage" --- that's sure nice of Mercedes to inform Q4 2016 Tesla buyers what hardware they are getting!

 

[MarkS22]

Regardless of whether that is true or not, perception often trumps reality, and Mercedes is telling this to the public... I doubt Tesla will sit back and let this continue to be true for very long...

View attachment 195449 View attachment 195450

Yeah, my other car is a 2016 Mercedes, fully loaded with all the driver assistance tools. It has all of that: stereo cameras, extra radars, you name it. I can tell you without hesitation, the extra sensors are useless without the software and infrastructure (OTA updates, fleet learning, etc) to support it.

I would hope Tesla isn't going to add sensors solely because Mercedes is bragging about them, public perception or not. It's not about "sitting back." They're working very, very hard on this. I'm not sure why everyone is so concerned about sensors while new improvements are literally being pushed out tonight. If the car's capabilities were stalled and not moving forward, I'd be in total agreement.

 

[KZKZ]

Yeah, my other car is a 2016 Mercedes, fully loaded with all the driver assistance tools. It has all of that: stereo cameras, extra radars, you name it. I can tell you without hesitation, the extra sensors are useless without the software and infrastructure (OTA updates, fleet learning, etc) to support it.

I would hope Tesla isn't going to add sensors solely because Mercedes is bragging about them, public perception or not. It's not about "sitting back." They're working very, very hard on this. I'm not sure why everyone is so concerned about sensors while new improvements are literally being pushed out tonight. If the car's capabilities were stalled and not moving forward, I'd be in total agreement.

The extra sensors are hardly useless.

The Blind Spot detection is light years beyond what Tesla offers. Tesla is basically blind to the rear.
The ability to detect and brake for stopped vehicles is not useless, as we've seen.
Crossing traffic detection is not useless, as we've seen.
Auto Lane change safely while automatically detecting fast approaching cars is not useless.

Tesla tried to use a very basic camera system and turn it into something it was never meant to do. The fact they are now just trying to add the radar into the mix tells volumes about how rushed and poorly designed the initial design was.

The results speak for themselves.

 

[KZKZ]

Regardless of whether that is true or not, perception often trumps reality, and Mercedes is telling this to the public... I doubt Tesla will sit back and let this continue to be true for very long...

View attachment 195449 View attachment 195450

Mercedes has already moved well beyond these 2014 driver assist features.

 

[Canuck]

I'm not sure why everyone is so concerned about sensors while new improvements are literally being pushed out tonight. If the car's capabilities were stalled and not moving forward, I'd be in total agreement.

Because the improvements being pushed out tonight make no difference to my vehicle, which I got just before AP hardware was added, and I'm not going to upgrade now only to find out that my new vehicle rolls off the line just before AP2.0 hardware comes out.

So you can call me bias because it's true.

 

[987S4]

I don't think AP is coming soon for many of the reasons listed already, but the one sign that gives me pause is the huge Q3 clear out that's going on right now.

I understand that it's supposed to make Q3 numbers look good, but from my vantage point they've gone overboard with the deals. Will an unprecedented discount and push of cars to achieve artificial numbers really make a difference for another capital raise? If not, then the only other reason for such a huge clearout would be amazing new hardware in Q4 that will continue to make them demand constrained moving forward.

 

[Canuck]

I don't think AP is coming soon for many of the reasons listed already, but the one sign that gives me pause is the huge Q3 clear out that's going on right now.

I understand that it's supposed to make Q3 numbers look good, but from my vantage point they've gone overboard with the deals. Will an unprecedented discount and push of cars to achieve artificial numbers really make a difference for another capital raise? If not, then the only other reason for such a huge clearout would be amazing new hardware in Q4 that will continue to make them demand constrained moving forward.

Are you not arguing against yourself? If the huge clear-out push might be because of amazing new hardware in Q4, then why do you also not think new AP hardware is coming soon? Those points seem contradictory to me.

 

[987S4]

Yup, I am arguing with myself.

I can't see them releasing new hardware until they can assess the new information that comes with the 8.0 update and the extra information the radar is providing.

However, I also don't understand the Q3 push. It seems extraordinarily desperate, and Tesla doesn't seem like it needs to be desperate. It's the one thing that makes me think Tesla has something up their sleeve for Q4.

If I had to bet one way or the other, I'd still bet that new AP hardware is a year out, but obviously I'm just a random internet guy speculating because I have nothing better to do.