Firmware 8.1 - Autopilot HW2

[bswn1]

It's interesting they're reportedly using two cameras. I don't think this is particularly good news given they couldn't recreate the Mobileye level of functionality with a single camera. Hopefully they're only using it to smooth the lane lines and there will be a lot more functionality to unlock soon.

 

[Todd Burch]

It's interesting they're reportedly using two cameras. I don't think this is particularly good news given they couldn't recreate the Mobileye level of functionality with a single camera. Hopefully they're only using it to smooth the lane lines and there will be a lot more functionality to unlock soon.

First, the former does not imply the latter. Just because they're using 2 cameras does NOT mean they couldn't get the level of functionality that mobileye has with one camera. It's really just that they're moving forward toward implementing all 8 cameras. If they turned on all 8 cameras in this release would you take that to mean that all 8 cameras were needed to keep the car in its lane?

On the contrary, in around a year or so they've managed to recreate most of the capabilities of MobilEye's entire technology in work since 1999. Every other manufacturer needs to use Mobileye or develop their own vision technology. Tesla now has a huge advantage in that they own their own vision processing technology.

 

[bswn1]

First, the former does not imply the latter. Just because they're using 2 cameras does NOT mean they couldn't get the level of functionality that mobileye has with one camera. It's really just that they're moving forward toward implementing all 8 cameras. If they turned on all 8 cameras in this release would you take that to mean that all 8 cameras were needed to keep the car in its lane?

On the contrary, in around a year or so they've managed to recreate most of the capabilities of MobilEye's entire technology in work since 1999. Every other manufacturer needs to use Mobileye or develop their own vision technology. Tesla now has a huge advantage in that they own their own vision processing technology.

I think it's abundantly clear this roll out did not go as planned. They expected parity three months ago. MONTHS! They activated the second camera out of necessity just to get the system to parity, and we aren't even there yet. You think there's no implications here?

 

[charliev7]

With 8.1, my autosteer is working well up to 80 mph but my auto lane change is not working. No cars in the desired lane in front or behind me but when I put the blinker on, the lines stay solid and the car stays put. Anyone else with the same problem?

 

[dcclappers]

With 8.1, my autosteer is working well up to 80 mph but my auto lane change is not working. No cars in the desired lane in front or behind me but when I put the blinker on, the lines stay solid and the car stays put. Anyone else with the same problem?

The first time I attempted it had the same problem. Tried again a minute or two later and it completed the auto lane change.

 

[mitchellh3]

Following up on my experiences. I drove about 15 miles on the highway last night at 80 mph (full speed!) and it performed extremely well. It still "ping pongs" in my opinion (my parents described it as feeling like a slightly "drunk" driver) but it stayed in its lane and I was able to go from on ramp to off ramp without manually steering. I traversed all lanes via auto lane change.

Weirdly I felt it was better at seeing the lanes at night. My theory there is just the contrast is much better at night, since I personally as a human also felt it was easier to see the lane markings at night. Great!

 

[t3slaMS]

I tried the auto lane change and it is perfectly smooth at lower speeds (<45mph) but at high speeds it does not take into account it is going fast and just dives into the new lane rather than adjusting the angle of turn and making it smooth.

 

[stopcrazypp]

I think it's abundantly clear this roll out did not go as planned. They expected parity three months ago. MONTHS! They activated the second camera out of necessity just to get the system to parity, and we aren't even there yet. You think there's no implications here?

We already know why Tesla gave that expectation 3 months ago. Tesla had planned to have the cars still have the Mobileye chip for the transition and then slowly transition over to Tesla Vision. Mobileye of course refused to allow this, so Tesla had to go on its own.

But as noted already, promises aside, reaching close to parity in 5 months after unveiling (whether using one or two cameras) is still mighty impressive (although I should note, that obviously behind the scenes Tesla had been working on Tesla Vision for far longer). Mobileye's chip is from more than a decade's worth of continual development.

 

[Deans]

With 8.1, my autosteer is working well up to 80 mph but my auto lane change is not working. No cars in the desired lane in front or behind me but when I put the blinker on, the lines stay solid and the car stays put. Anyone else with the same problem?

Did you remember activate Auto Land Change in the settings menu?

 

[chillaban]

But as noted already, promises aside, reaching close to parity in 5 months after unveiling (whether using one or two cameras) is still mighty impressive (although I should note, that obviously behind the scenes Tesla had been working on Tesla Vision for far longer). Mobileye's chip is from more than a decade's worth of continual development.

Agreed 100% with this. Tesla is squarely wrong for their utterly unrealistic and borderline misleading initial promises of parity (and depending on how you read it, all the EAP features) in December. Whether or not one feels personally misled by this, I think we can all agree it's not a great way of doing business and retaining customer trust.

However on the tech side, the fact that right now, 8.1 AP2 5 months later after announcement and fewer months after initial product ship, is starting to look as good as AP1 was at launch.... that's very impressive and shows technical competency at Tesla in their Autopilot program. Full stop. It's becoming more and more clear why Tesla and MobileEye did not work well with each other -- the pace that Tesla is trying to drive is fairly aggressive and MobileEye's slow-and-steady approach was no longer cutting it. They've talked about EyeQ4 before the launch of the product for longer than it took for Tesla to replicate EyeQ3.

In the end though, I think competition is very good in this field. Tech companies have a way of resting on their laurels unless a viable competitor shows up. Now we have two (three if you count Mercedes's in-house systems) serious contenders in the path to autonomy on production cars.

 

[StormShadowX]

I updated to 8.1 last night and took it for a drive on the 101 freeway this morning. It didn't seem like there was calibration needed as I was able to immediately activate Autosteer and drive at about 70-75 mph. What i noticed immediately when I turned on AS was that it quickly tried to align itself toward the right lane... sometimes even driving directly onto the right lane marking for a quick second before re-adjusting itself back onto the lane, but never exactly able to get it to be centered. It was always driving too close to the right lane marking, leaving itself a bit of a gap from the left lane marking. I turned AS off and on, and repeated this several times, and every single time it was doing exactly the same thing. This was particularly an issue for me when driving next to a big truck and because it was still hugging the right lane marking too closely that I had to just cancel AS several times to retake control. Is anyone else experiencing this with the new 8.1 update?

 

[OBX John]

Unless I'm missing something, with Mobileye valued at 15 Billion by Intel, Tesla basically increased their own company value by 15 Billion with less than a year of self driving development. Pretty impressive. And now since they are the only one with cars in the field, that value may keep increasing since competitors have yet to field cars with equivalent systems so Tesla will maintain a first mover advantage. Perhaps Mobileye doesn't need a fleet of thousands of cars to continue development at the same pace, but I'm skeptical.

 

[chillaban]

Unless I'm missing something, with Mobileye valued at 15 Billion by Intel, Tesla basically increased their own company value by 15 Billion with less than a year of self driving development. Pretty impressive. And now since they are the only one with cars in the field, that value may keep increasing since competitors have yet to field cars with equivalent systems so Tesla will maintain a first mover advantage. Perhaps Mobileye doesn't need a fleet of thousands of cars to continue development at the same pace, but I'm skeptical.

Well, the MobileEye CTO is very good at talking with a set of slides and leading people to believe they are capable of doing a lot in the near future.

I would say MobileEye still has a valid autonomy approach and while Tesla is very close to an impressive game of catch-up, we still have yet to see Tesla eclipse MobileEye in capability and future product roadmap.


(P.S. Quite frankly, the $15 bil value of MobileEye is how they use regulatory incentives force their cameras down BMW/VW/etc's throats because it buys free EURO NCAP points)

 

[TexLaw]

Did you remember activate Auto Land Change in the settings menu?

That was my first thought, too, just because I forgot to turn it on myself.

 

[TexLaw]

I think it's abundantly clear this roll out did not go as planned. They expected parity three months ago. MONTHS! They activated the second camera out of necessity just to get the system to parity, and we aren't even there yet. You think there's no implications here?

I dare say the strongest implication is that if someone gave you a bar of gold, you'd complain that it was heavy.

A few months' delay on a software or firmware rollout is virtually nothing. I hate to even imagine how much of that delay was due to regulatory approval, which tends to move at glacial speed. This really is terra incognita, and any timeline estimates necessarily include a heavy dose of speculation.

I got to give autosteer and lane change a good try at 65mph, earlier today, and I was quite impressed. The car held the lane much more smoothly, and the lane change was much smoother than I expected from earlier reports on TMC. If that's because two cameras are working, then so be it.

 

[Tangible]

I notice there are lots of Sunbelt folks commenting here. The climate here in the Northeast tends to leave our highways (we don't have Freeways; EZ Pass sees to that) with faded lane markings at the best of times, and snow-covered ones in the wake of storms. Are there any other hardy New Englanders or midwesterners on this forum. If so, how are the AS features working for you?

 

[Todd Burch]

I think it's abundantly clear this roll out did not go as planned. They expected parity three months ago. MONTHS! They activated the second camera out of necessity just to get the system to parity, and we aren't even there yet. You think there's no implications here?

I never said it went as planned. Obviously it was planned to be released months ago, yes. But things happen. Like, say, your Director of Autopilot leaves the company (possibly stealing data) to go start his own self-driving company. That could have an impact. Plus, Elon tends to be a bit optimistic.

Regardless, I'm not sure how that's relevant to your statement that because 2 cameras are now turned on, 2 cameras are needed to keep the car in its lane to match AP1.

"They activated the second camera out of necessity just to get the system to parity" is absolute conjecture supported by zero evidence.

Machine learning is a process that requires lots of data. When AP1 came out it was pretty shaky at times and dove for the exits. It took 2 years of refinement to get where it is now. Doesn't mean the MobilEye camera was insufficient, as evidenced by how stable AP1 is now (as it has matured).

 

[bswn1]

I dare say the strongest implication is that if someone gave you a bar of gold, you'd complain that it was heavy.

A few months' delay on a software or firmware rollout is virtually nothing. I hate to even imagine how much of that delay was due to regulatory approval, which tends to move at glacial speed. This really is terra incognita, and any timeline estimates necessarily include a heavy dose of speculation.

I got to give autosteer and lane change a good try at 65mph, earlier today, and I was quite impressed. The car held the lane much more smoothly, and the lane change was much smoother than I expected from earlier reports on TMC. If that's because two cameras are working, then so be it.

I'm not complaining at all. I'm on my second car and have two 3s reserved. What's happened to this community that we can't dare discuss what's clearly right in front of our faces without fear of being called a complainer?

 

[croman]

I dare say the strongest implication is that if someone gave you a bar of gold, you'd complain that it was heavy.

A few months' delay on a software or firmware rollout is virtually nothing. I hate to even imagine how much of that delay was due to regulatory approval, which tends to move at glacial speed. This really is terra incognita, and any timeline estimates necessarily include a heavy dose of speculation.

I got to give autosteer and lane change a good try at 65mph, earlier today, and I was quite impressed. The car held the lane much more smoothly, and the lane change was much smoother than I expected from earlier reports on TMC. If that's because two cameras are working, then so be it.

Don't push a regulatory delay theory. It's completely untrue. Regulations are part and parcel of civilized society and are scapegoated too easily.

 

[ucmndd]

Updated last night, first experience this morning was good. Completed 30 miles of freeway driving at 80mph, multiple lane changes, without any real issues and no need to interrupt or take over. One lane change was a little jerky, the rest were very smooth. Very pleased with the progress.