Blog WSJ: Ambitious Autopilot Push Angered Tesla Engineers

Tesla engineers are reportedly jumping ship because they don’t believe Autopilot 2.0 hardware can meet the fully self-driving promise that Elon Musk proudly touts.

The Wall Street Journal says Sterling Anderson, previously the Autopilot director, decided to leave Tesla in December in part because he didn’t agree with the claims Musk was making about the vehicle’s potential for full autonomy.

According to the WSJ (paywall):

In a meeting after the announcement, someone asked Autopilot director Sterling Anderson how Tesla could brand the product “Full Self-Driving,” several employees recall. “This was Elon’s decision,” they said he responded. Two months later, Mr. Anderson resigned.

The Autopilot division has lost some 10 employees and four managers recently, according to the report. Satish Jeyachandran, the former director of hardware engineering for Tesla’s Autopilot team, and Berta Rodriguez-Hervas, a former machine learning manager also left the company in June. Anderson was succeeded by Chris Lattner, a former Apple developer, but he left in June after just six months on the job.

Tesla has declined to comment on the report.

 

[bonnie]

I exceeded 9003 before 1994. Some of their concepts come from my white papers. As does ISO 9001 and AS 9100. And D6-51991, digital mfr'g and inspection.

If Tesla Motors was in a loop of continuous improvement or even used preventative action systems, WTF happened to AP2???

Haha, fair question.

I admit, though I first brought a company thru 9001 in the very late 80s, I am much more intimately familiar with medical device standards like 13485 and risk management (14971). I contributed to the latter and have taught several courses on both. There's a reason that FDA has not outsourced pharma and device company inspections to Notified Bodies here in the States - simply too many examples of companies that don't even come close to meeting even the intent of the standards, yet they have a pretty cert framed in the front lobby from what I thought was a reputable Notified Body.

 

[McRat]

Haha, fair question.

I admit, though I first brought a company thru 9001 in the very late 80s, I am much more intimately familiar with medical device standards like 13485 and risk management (14971). I contributed to the latter and have taught several courses on both. There's a reason that FDA has not outsourced pharma and device company inspections to Notified Bodies here in the States - simply too many examples of companies that don't even come close to meeting even the intent of the standards, yet they have a pretty cert framed in the front lobby from what I thought was a reputable Notified Body.

Unnamed major medical companies....

The FDA required them to have blueprints for all of their tooling.
They sent the tooling to us, and we made blueprints from the tools to ±0.002 mm.

The FDA said all medical device tooling must be inspected.
We inspected the tools to the blueprints we created from their tooling.

It is probably no big surprise that every single one of the 1,100+ tools were in print, verified by an ISO registered company. Imagine that.

 

[bonnie]

Unnamed major medical companies....

The FDA required them to have blueprints for all of their tooling.
They sent the tooling to us, and we made blueprints from the tools to ±0.002 mm.

The FDA said all medical device tooling must be inspected.
We inspected the tools to the blueprints we created from their tooling.

It is probably no big surprise that every single one of the 1,100+ tools were in print, verified by an ISO registered company. Imagine that.

The manufacturer was responsible for 1) ensuring that the tools were reproducible/documented, 2) that tolerances were validated to not negatively impact the product in any way (and unless they supplied you with the risk assessment, you couldn't have done that), and, 3) that the person or company doing the inspection meet basic quality standards.

I would have at least asked you for the training records of the person doing the inspection, insisted that they not be the person who did the drawings, and wanted to see your calibration program for the tools used to measure. Nothing personal . FDA is much tougher than any Notified Body. They bump up the evidence bar considerably - someone inspecting for ISO compliance is actually assisting the business. An FDA inspector is a federal law enforcement agent & they are trained to collect evidence suitable for use in court.

Nice to know this side of you.

--- Edit: You're kidding - they had 1100+ tools without documentation??? I'd say they were dealing with a Warning Letter at that point. That's inexcusable.

 

[McRat]

My personal experience with a number of notified bodies is that the certification may or may not mean anything. ...

Registrars are like a box of chocs Forrest. When you do a site survey and find an a Major on a company that passed 9001, it is your duty to report them to ISO.

 

[whitex]

There may be engineers who disagree with Tesla because they think what Tesla promise is impossible or unethical but Tesla has proven them wrong.

Please read the above example when Panasonic thought that Tesla was crazy to ask it producing more, not just more, but much more cells. It thought it's impossible to have that much of demand. It's been in the business along time and it is the expert of how much cells it should produce but Tesla was talking about a Gigafactory? It's just impossible.

Tesla sold Panasonic the idea and now Panasonic has admitted that it was wrong!

An important distinction you are not making, but should is taking money for something. For example, if a drug company says they working on a cure for cancer and believe they'll have it soon, they are hero's. If instead they start taking patients' money for a cure they don't have, they are villains. See the difference?

 

[ecarfan]

Taking money for something impossible to deliver is not unethical? You seems to have very unique set of ethics.

Your critique is based on a false assumption. No one has demonstrated that FSDC is "impossible to deliver". That is an opinion. It may be correct, or it may not be correct. We don't know yet.

 

[alcibiades]

No. I drive a beat up Honda.

Tesla's product descriptions have made it clear to me that the AP2 software was incomplete at launch and would gradually improve over time.

Two things.

1. That's what I figured.

2. You don't know what you're talking about.

 

[Tam]

...See the difference?...

We keep going over and over the same point over and over again and again:

Tesla does not force any body to pay for Self-driving.

People have a choice.

If they want to, they can read what they pay for:

Self-driving only works after:

1) validations
2) regulatory approvals

with no timeline mentioned.

So, be informed and don't just blindly pay for it then complain it's a beta and I'm a Guinea pig tester, I've been conned, I'm a victim..

 

[alcibiades]

We keep going over and over the same point over and over again and again:

Tesla does not force any body to pay for Self-driving.

People have a choice.

If they want to, they can read what they pay for:

Self-driving only works after:

1) validations
2) regulatory approvals

with no timeline mentioned.

So, be informed and don't just blindly pay for it and complain it's a beta and I'm a Guinea pig tester.

FSD is only released after "validation" (snort) and regulatory approvals (snort).

They are free to demonstrate it and produce videos at any time. They haven't. There has been significant turmoil in the AP group. There was an unexplained and false 3-6 month timeline for feature introduction given by the CEO. Before any code has been released to enable any functions for even EAP (expected to complete validation in 12/16), Tesla is making significant hardware changes.

Forget the timeline of when it will be completed and enabled on my car; it could take some time -- that's ok. But let's just see any HW2 car demo shuttle people like Cruise and Waymo. Let's see an EAP car demonstrate the automatic lane changes, the transitions from one highway to another, and the exiting of a freeway when the destination is near.

 

[bonnie]

Two things.

1. That's what I figured.

2. You don't know what you're talking about.
View attachment 243948

It was clear to me and I have AP2. (Not sure why someone needs to drive it to be able to comment, but hey, here I am.) I get that people are upset

When you have a blueprint for making a medical device (or pretty much anything), you'll notice that the name of the person who created the B/P and date is listed. As is the person who checked it.

When you inspect a medical device or pretty much anything, you'll notice the inspection report includes the inspection tech and date is on the report per ISO 17025.

You are not allowed validate your own work per ISO 9001 or 17025 when you can benefit from the results matching.

But both require validation from a minimum of 3rd source which is management.

When you create B/P's to ±0.002mm and your machines can resolve to 4 to 8 times that level, the the B/P's is normally going to match the inspection results regardless of which machine was used, especially when you allow a tolerance window 10 times the B/P dimension.

Note: The DOE is a few levels past the FDA, trust me on this.

I know the regs well, trust me . (Have trained more than a few FDA agents on medical device ... not to mention working with FDA and various ISO orgs on harmonization.)

No, medical device regs do not require validation from a minimum of a 3rd source. What the regs do require (both US 21 CFR Part 820 and ISO 13485) is that you are able to demonstrate independence and that the level of testing is commensurate with perceived risk to end users.

DOE is not a few levels past - but agree that it is very different. They don't use the same risk-based approach, but they do require additional eyes on the validation work. It is much more proscriptive in approach, which means you have very definitive guidelines to apply. My common explanation with FDA's approach is that 'If they tell you you need to be in Tucson by Tuesday at 10am, that's when you need to be there. They won't tell you how you should travel. Bus, train, hitchhike, drive, whatever. But if you're late you don't get to blame it on the transport you chose. You chose it. You're accountable for arriving on time.' Definitely a different approach and for those who don't understand the regulations in depth, it's much more confusing than DOE which is more a 'one size fits all' approach. But I'd argue better results with FDA's approach.

Tolerances are a tricky area. If a good stackup analysis is not done, I've found that engineering is overly conservative regarding tolerances just so they can pass testing. And that leads to rejection in I/I, since parts with tighter tolerances are more expensive to make. And then engineers try dispositioning as Use As Is and you end up with crappy product in the field. So I don't much care what the tolerances are, but rather how the tolerances were determined.

 

[Krugerrand]

Which Registrar?
Pretty much all high level mfr'g organizations are ISO 9001:2008 or 2015.

I searched hard and could not find any indication that Tesla Motors can prove they have a Quality Control System of any kind.

Took me 5 seconds to find reference to it: Quality assurance and Quality management | Tesla

 

[alcibiades]

It was clear to me and I have AP2. (Not sure why someone needs to drive it to be able to comment, but hey, here I am.) I get that people are upset

Oh hey, cool.

 

[Krugerrand]

It will be on the Tesla website. Show me. List the registrar and I will call.

I don't have to do anything. If it's that important to you, try a little harder than you tried before.

 

[anticitizen13.7]

Two things.

1. That's what I figured.

2. You don't know what you're talking about.
View attachment 243948

Can anyone verify the source and validity of the screen shot? And how long that version of the page was up?

Tesla's official announcement on Oct 19, 2016, did not place a timeline on software feature delivery: All Tesla Cars Being Produced Now Have Full Self-Driving Hardware

Teslas with new hardware will temporarily lack certain features currently available on Teslas with first-generation Autopilot hardware, including some standard safety features such as automatic emergency braking, collision warning, lane holding and active cruise control. As these features are robustly validated we will enable them over the air, together with a rapidly expanding set of entirely new features.

Given that Autopilot 1 features had taken at least a year to mature, I never believed that Autopilot 2 would be fully ready between Oct 19 and Dec 31 of 2016.

Perhaps I'm too familiar with software development uncertainties while the general public is not.

 

[bonnie]

Though I don't know what certs Tesla has, I took a quick look at some of the job descriptions and expectations. @McRat Since one of the job duties is leading ISO certification, I will take that to mean that either they are getting it for the first time OR they are expanding current certs.

Sr. Quality Engineer | Tesla | Tesla

 

[whitex]

Your critique is based on a false assumption. No one has demonstrated that FSDC is "impossible to deliver". That is an opinion. It may be correct, or it may not be correct. We don't know yet.

No, my critique was based on an explicit statement made by Tam stating that it is not unethical for Tesla to sell something impossible to deliver.

Your critique of my comment seems to be based on incomplete reading and assumption of context.

 

[oktane]

No, my critique was based on an explicit statement made by Tam stating that it is not unethical for Tesla to sell something impossible to deliver.

Your critique of my comment seems to be based on incomplete reading and assumption of context.

No one has demonstrated that it's possible to deliver either, what's your point? Every day that goes by without parity for AP2 makes it less likely that they will ever reach parity or deliver FSD. That is obvious.

 

[Matias]

Can anyone verify the source and validity of the screen shot? And how long that version of the page was up?
.

It was up at least till January 12th 2017. So anyone who purchased EAP before that had a promise from Tesla, that it is only subject to regulatory approval, the code itself was ready...

Order a Tesla Model X | Tesla

OMG, NOW I UNDERSTAND IT! Because in January 12th 2017 Tesla expected the code to be ready in December 2016, the only possible explanation is, that Tesla is developing a time machine and by using it, they will later in the future travel back to December 2016 and present that code for regulatory approval. It all makes sense now!

 

[alcibiades]

Can anyone verify the source and validity of the screen shot? And how long that version of the page was up?

That's exactly the reaction that is going to win the consumers the EAP suit. In August of 2017 it is viewed as unbelievable that Tesla would have said what they did to quarter 4 2016 buyers.

 

[oktane]

That's exactly the reaction that is going to win the consumers the EAP suit. In August of 2017 it is viewed as unbelievable that Tesla would have said what they did to quarter 4 2016 buyers.

The case is a slam dunk. Bad news for Tesla.