Experts ONLY: comment on whether vision+forward radar is sufficient for Level 5?

[Joe F]

I can build a better implementation of ap2 in a few days.

Now that's really great news! Please get back to us when you start your own car company, and implement your own Auto Pilot system. I'm sure it's going to be the best, ever!

I for one can't wait to sell my Tesla any buy your best in class car!

 

[JeffK]

I'm not talking about some demo that was based on nvidia driveworks.

For the millionth time, they are not using Driveworks.

Tesla having the best eyeq3 implementation in the industry is another myth

Myth according to you, but not a myth according to professional reviewers such as Motor Trend or Car and Driver

 

[Bladerskb]

For the millionth time, they are not using Driveworks.


Myth according to you, but not a myth according to professional reviewers such as Motor Trend or Car and Driver

yes they used driveworks for that specific video. other than that we don't know specific features they are adopting from driveworks. As driveworks is open source they can literally use anything, even if its just one line of code. But something tells me they will be using a lot more including the need to optimize their code by following some of driveworks coding standards.

what review? from 2015? from 2016? without Volvo? your kidding right?
the tests so far, not only have they not included Volvo but they are filled with cognitive bias and placebo effects.

 

[stopcrazypp]

yes they used driveworks for that specific video. other than that we don't know specific features they are adopting from driveworks. As driveworks is open source they can literally use anything, even if its just one line of code. But something tells me they will be using a lot more including the need to optimize their code by following some of driveworks coding standards.

what review? from 2015? from 2016? without Volvo? your kidding right?
the tests so far, not only have they not included Volvo but they are filled with cognitive bias and placebo effects.

Volvo's system was a joke when the reviews came out (limited to very low speeds). The Mercedes system was the state of the art then (other than Tesla).
He's talking about these reviews:
Semi-Autonomous Cars Compared! Tesla Model S vs. BMW 750i, Infiniti Q50S, and Mercedes-Benz S65 AMG - Feature
http://www.motortrend.com/news/testing-semi-autonomous-cars-tesla-cadillac-hyundai-mercedes/

As for Volvo's newer Pilot Assist 2 (PA2) system, reading the Volvo threads about them, there are the same problems.
Auto Pilot Assist 2 thread (experience, problems)

 

[Bladerskb]

Volvo's system was a joke when the reviews came out (limited to very low speeds). The Mercedes system was the state of the art then (other than Tesla).
He's talking about these reviews:
Semi-Autonomous Cars Compared! Tesla Model S vs. BMW 750i, Infiniti Q50S, and Mercedes-Benz S65 AMG - Feature
Testing (Semi) Autonomous Cars With Tesla, Cadillac, Hyundai, and Mercedes - Motor Trend

As for Volvo's newer Pilot Assist 2 (PA2) system, reading the Volvo threads about them, there are the same problems.
Auto Pilot Assist 2 thread (experience, problems)

I don't see Volvo in any of those reviews.

and do you want me to link the thousands of ap1/ap2 problem threads?
sure it fails in very steep/sharp curves, just as ap1 does.
Right now Volvo Pilot Assist 2 > AP1 > AP2.
You can't dispute this.

 

[croman]

Another myth is tesla's supposed lead in tech and horde of IP.
But when you actually take a look it, tesla has the least amount of patents among car makers, infact. Other car makers patent more tech a year than tesla has in total.

Mobileye for example has a horde of patents relating to driverless cars and vision. you think they won't come for tesla when the time is right?

Patents by Assignee Mobileye Vision Technologies Ltd. - Justia Patents Search

here is mobileye 8 camera system that tesla copied. including the exact placement.
36mins 24secs

If you knew anything about patents you'd know its not the sheer number of patents in a portfolio but rather how they use them, right? The right combo can be quite pleasing.

 

[Bladerskb]

If you knew anything about patents you'd know its not the sheer number of patents in a portfolio but rather how they use them, right? The right combo can be quite pleasing.

tesla has NONE, ZERO, ZIP, 0 as it relates to autonomous cars.

Google for example has 300 as of 2015 and mobileye has hundreds.

 

[JeffK]

yes they used driveworks for that specific video.

Source?

 

[croman]

tesla has NONE, ZERO, ZIP, 0 as it relates to autonomous cars.

Google for example has 300 as of 2015 and mobileye has hundreds.

Tesla is leading from behind.

 

[Bladerskb]

As for Volvo's newer Pilot Assist 2 (PA2) system, reading the Volvo threads about them, there are the same problems.
Auto Pilot Assist 2 thread (experience, problems)

infact the person who made that thread later on made this post.

I am happy that today I found very easy way how to "upgrade" it to PA III

The result is, I am now able to let the PA to drive by itself all the time without those annoying requests. I did about 200 km on Swiss highways today, touching the steering wheel maybe thrice ... (when the curve was too steep). Finally it works as it should by one slightly funny adjustment

That's 125 miles with only 3 interventions and those were only on very steep curve.

A lot of the complaints seem to rest on the 15 seconds interval and very steep curves.
But in tesla forums we have regular false positive/negative hard brakings, not seeing the car ahead and not braking, hundreds of accidents, truck lust, tons of rear end and near rear ends, etc. So on and so forth.

 

[stopcrazypp]

I don't see Volvo in any of those reviews.

Sorry, I didn't make it clear. The Volvo system was a joke when those articles came out (didn't even work at highway speeds), so they didn't even bother reviewing it.

and do you want me to link the thousands of ap1/ap2 problem threads?
sure it fails in very steep/sharp curves, just as ap1 does.
Right now Volvo Pilot Assist 2 > AP1 > AP2.
You can't dispute this.

Until we get objective comparison reviews of the system (similar to the linked articles where vehicles are tested under same conditions), this is just speculation.

 

[bob_p]

I can build a better implementation of ap2 in a few days. In fact I don't have to, geohut already has. His openpilot 0.3 is better than ap2. That's just one guy in a couple weeks. You bringing up Mobileye and saying 18 years shows how naive you are. The technology and the scientific breakthrough 18 years ago pales in comparison to today. I can do something in 1 hour that it takes me a decade to do 18 years old.

It's clear Tesla underestimated how much development would be required to get Tesla Vision to provide object recognition as well as Mobileye-based AP1. Since the AP2 sensors should be an improvement over the sensors in AP1, it should only be a matter of time before Tesla is able to get the new Vision software operating well enough to get the AP2 functionality working as well as AP1.

Underestimating the amount of time required to get software operational is not uncommon - Tesla isn't alone in suffering from this optimism, as is evidenced by the above quote.

 

[JeffK]

It's clear Tesla underestimated how much development would be required to get Tesla Vision to provide object recognition as well as Mobileye-based AP1.

We did learn this wasn't necessarily their fault as they had originally planned to incorporate the MobilEye EyeQ3 chip in the board until the transition to the actual AP 2.0 software they've been working on before the whole very public break up.

 

[bob_p]

The change away from MobilEye was known BEFORE Tesla announced AP2 was going into production.

They were consistent at announcement, and even into early 2017, continued to claim on the Tesla website the AP2 hardware would be operating at AP1 level by December 2016.

I agree that if they had included the MobileEye hardware in AP2, they would have been able to achieve AP1 level faster - but by the time Tesla announced AP2, they already knew they had to develop Tesla Vision to replace MobileEye - and they considerably underestimated how long it would take.

Tesla should have known in November (or possibly October) that they weren't progressing quickly enough to achieve their December goal - and they didn't change the wording on the Tesla website until early 2017. And with their lack of transparency on the schedule and status of getting AP2 operating at AP1 levels, they created an opening for the lawsuit.

I've managed large projects with aggressive schedules and can empathize with the software team. They likely faced incredible pressure to get Tesla Vision working as quickly as possible - and probably saw there was a path, if everything fell into place, where they could achieve the December goal. But, projects like this are always challenging, things rarely work out as well as you expect, and it usually takes developers longer to get things working than they planned.

Unfortunately, this follows the same pattern we've seen from Tesla since the Model S was produced. The public announcements by Musk and Tesla tend to be optimistic, promising functionality that is delivered late (or not at all, like the long-promised browser improvements) and when we do get the releases, they often have major bugs or design flaws (like the missing time-of-day clock in 7.0 or multiple media player issues that are still present over 6 months after 8.0 was released).

As Tesla starts ramping up Model 3 production and shifts away from mostly serving "early adopters", they risk customer disappointment if they continue to deliver promised functionality late - and with bugs/flaws...

 

[JeffK]

The change away from MobilEye was known BEFORE Tesla announced AP2 was going into production.

They were consistent at announcement, and even into early 2017, continued to claim on the Tesla website the AP2 hardware would be operating at AP1 level by December 2016.

I agree that if they had included the MobileEye hardware in AP2, they would have been able to achieve AP1 level faster - but by the time Tesla announced AP2, they already knew they had to develop Tesla Vision to replace MobileEye - and they considerably underestimated how long it would take.

Tesla should have known in November (or possibly October) that they weren't progressing quickly enough to achieve their December goal - and they didn't change the wording on the Tesla website until early 2017. And with their lack of transparency on the schedule and status of getting AP2 operating at AP1 levels, they created an opening for the lawsuit.

I've managed large projects with aggressive schedules and can empathize with the software team. They likely faced incredible pressure to get Tesla Vision working as quickly as possible - and probably saw there was a path, if everything fell into place, where they could achieve the December goal. But, projects like this are always challenging, things rarely work out as well as you expect, and it usually takes developers longer to get things working than they planned.

Unfortunately, this follows the same pattern we've seen from Tesla since the Model S was produced. The public announcements by Musk and Tesla tend to be optimistic, promising functionality that is delivered late (or not at all, like the long-promised browser improvements) and when we do get the releases, they often have major bugs or design flaws (like the missing time-of-day clock in 7.0 or multiple media player issues that are still present over 6 months after 8.0 was released).

As Tesla starts ramping up Model 3 production and shifts away from mostly serving "early adopters", they risk customer disappointment if they continue to deliver promised functionality late - and with bugs/flaws...

Although it might have been known before production, it probably wasn't known before the planning phase.

Planning, design, board fabrication, validation, production, and then the need for fleet data meaning you actually need cars on the road. It's a huge undertaking in a short amount of time as you've mentioned, but the software team can't do it alone.

Anyone who follows Elon knows there wasn't a lack of transparency. Elon gave regular updates regarding the rollout of initial AP 2.0 features. The first of which, to a very limited set of vehicles, went out Dec 31st.

 

[electracity]

They likely faced incredible pressure to get Tesla Vision working as quickly as possible - and probably saw there was a path, if everything fell into place, where they could achieve the December goal.

The problem is that the edge cases are largely unseen until the solution for a certain problem is tested. This isn't straightforward application programming where good design reliably encompasses the structure needed to meet the primary goals. Where Tesla's development process is particularly ugly is that they have not ensured highly reliable ADAS safety functionality as the base on which to build higher self driving functions. I find Tesla's approach here egotistical and unethical.

 

[bob_p]

As someone who has managed large, aggressive software projects, I know it is possible to implement large systems, adding major new functionality, hitting aggressive schedules - and still avoid having major bugs & design flaws in the released software. It's possible, but requires a combination of competent & dedicated developers, well defined process & methodology, and strong leadership team.

Unfortunately, Tesla's software track record since the first Model S was produced hasn't been very encouraging. Releases are typically late. Long promised functionality still hasn't been provided. And releases still go through the start-stop release process as Tesla finds major bugs that were missed during their beta testing (using a small number of hand selected beta testers, evidently ignoring the offers from those of us who have considerable software & testing experience).

Tesla's inability to get even the media player operating correctly does raise concerns that mission critical (and considerably more complex) EAP/FSD may suffer similar issues as the software releases we've seen up until now.

Since Tesla hasn't been very transparent on how they are actually developing their software, I hope they are investing more in the Autopilot/FSD software development teams and methodology, and are on a path to deliver higher qualify releases for EAP/FSD...

 

[JeffK]

As someone who has managed large, aggressive software projects, I know it is possible to implement large systems, adding major new functionality, hitting aggressive schedules - and still avoid having major bugs & design flaws in the released software.

You should work for Apple, Microsoft, Facebook, etc. because every one of them can't do this 100% of the time. They must not be as successful as you are

You feel that Tesla put as much though into the media player as into mission critical items??? Do even remotely think the same software developers did both? Do you think the media player is as rigorously tested and validated as mission critical software?

Tesla doesn't need to be transparent on how they are developing software and that hasn't stopped you from assuming they are incompetent. Tesla has been transparent about their release and testing process but you still seem to think it goes magically from a hand full of beta testers straight to 100% of the general public which is not the case at all, and has never been the case. Tesla uses a similar process to other tech companies especially with the notion of shadow mode.

Here's a good book for you to read The DevOps Handbook: How to Create World-Class Agility, Reliability, and Security in Technology Organizations: Gene Kim, Patrick Debois, John Willis, Jez Humble, John Allspaw: 9781942788003: Amazon.com: Books

 

[Topher]

In the end it really doesn't matter if Tesla's FSD is 2X or 10X better than a human.... Any better and you'd start saving lives.

Except as regards adoption. Most people think they are better than average drivers. Easy to convince yourself that 2x isn't better than you.

tesla has NONE, ZERO, ZIP, 0 as it relates to autonomous cars.

Tesla doesn't WANT any patents. Elon said patents are a lottery to win a lawsuit. They gave away the ones they had.

Thank you kindly.

 

[lunitiks]

Tesla doesn't WANT any patents.

Yet they KEEP filing them.