By on October 31, 2016

Comma One Rear

George Hotz announced he was cancelling the Comma One project last week in response to an information request from the National Highway Traffic Safety Administration. At first glance, this might appear to be a bit of government overreach. However, once you start digging into the letter, it’s apparent the questions are reasonable and easy to answer.

The main goal of the questionnaire is to assess the safety of the Comma One device. NHTSA set a deadline of November 10th to receive the response or Hotz would risk a $21,000 a day fine. Hotz claims that the letter was threatening.

Lets look at the questions in detail and see how they break down.

The letter starts out with an explanation of the facts that NHTSA currently has along with references to the Comma.ai blog. It ends the first page with the following emphasised statement:

We are concerned that your product would put the safety of your customers and other road users at risk. We strongly encourage you to delay selling or deploying your product on the public roadways unless and until you can ensure it is safe.

The statement clearly asks for the product to be delayed and does not prevent Hotz from selling it in the future. The letter continues by stating assertions Hotz makes about the nature of the product are not enough to validate the safety of it and finished with the deadline and fine amounts. The special order follows with a set of 15 questions, examined below.

1. Describe in detail how the comma one is installed in a vehicle and provide a copy of installation instructions for the comma one.

The first question is simple enough and asks for the installation instructions and an explanation of how the device is installed. For a device that is set to hit the market in less than a month, Hotz should surely have some installation instruction already written up and ready to send out.

2. Describe in detail that advanced driver assistance features of comma one, including how those features differ from the existing features of the vehicles in which the comma one is intended to be installed.

This requests a breakdown of the features and how they improve upon the products currently installed in the vehicles in which Comma.ai is targeting. Many of these features have already been explained by Hotz in his presentations, so this should be something that’s easy for him to put together. The second part of the question might be a bit harder to answer, but it’s still feasible since he stated his system is an aftermarket improvement.

3. Describe in detail how a vehicle driver uses the comma one and provide a copy of user instructions for the comma one.

Providing an answer for this question should be easy enough. A user manual, much like the instruction manual mentioned above, should already describe the details on how a user may engage the system with less than a month until product launch.

4. Provide a detailed description of the conditions under which you believe a vehicle equipped with comma one may operate safely. This description must include:

a. The types of roadways on which a vehicle equipped with comma one may operate safely;

b. The geographic area in which a vehicle equipped with comma one may operate safely;

c. The speed range in which a vehicle equipped with comma one may operate safely;

d. The traffic conditions in which a vehicle equipped with comma one may operate safely;

e. The environmental conditions such weather or time of day in which a vehicle equipped with comma one may operate safely; and

f. The amount and type of driver inputs necessary for a vehicle equipped with comma one to operate safely.

This one gets into a bit more detail but is reasonable enough and mostly asks for detail on whether this can be operated on all roads or if it’s more of a highway device. It also asks how external conditions will impact it and what the driver will need to in response.

5. Provide a detailed description of the basis for your response to Request No. 4, including a description of any testing or analysis to determine safe operating condition for a vehicle equipped with comma one.

This one asks for some analysis and proof for the answer to the question above. Comma.ai should be able to provide it from the logs of its test vehicle.

6. Describe the steps you have taken or plan to take to ensure the safe operation of a vehicle equipped with comma one, including but not limited to automated shutoff of comma one features and owner education.

The automated shutoff and hand-off is a big part of autonomous vehicle development, so it’s fair to ask this question. Hotz should be able to answer this easily as his hand-off procedure was recently praised by testers who’ve been inside the test vehicle.

7. Provide a list by make, model, model year or year of production of each vehicle for which you support or anticipate supporting use of the comma one.

I asked this question along with multiple others on this page in a request to Hotz after the TechCrunch presentation. I was able to find some basic information on the comma.ai blog, but this is a question that he should be able to answer on the spot without much analysis.

8. Describe in detail any steps you have taken to ensure that installation of the comma one in any supported vehicle does not have unintended consequences on the vehicle’s operation.

This is a good question since Hotz appeared to be pushing code in an ad-hoc manner over the CAN-BUS and was not cooperating with American Honda to verify this would not have unintended consequences. If he did in fact have supporting data showing why his device would not interrupt other operations of the vehicle, Hotz should be able to answer this question easily.

9. Describe the functionality of comma one, if any, if installed in an unsupported vehicle.

This is important since Hotz was planning to use an OBD II device that would plug into supported vehicles. Since this is a universal standard, the device could be plugged into any other vehicle and could cause unintended consequences since Hotz was forcing messages over the CAN-BUS.

10. Have you done any analysis or testing of the impact or potential impact of comma one on the vehicle’s compliance with FMVSS? If yes, please describe the analysis or testing in detail and provide supporting documentation. If no, describe why not.

Federal Motor Vehicle Safety Standards cover design, construction, performance, and durability requirements for motor vehicles and any safety-related components, systems, and design features. The Comma One falls into this bucket since it would be controlling braking and other safety functions. At this point in the launch cycle, Comma.ai should at least have a general overview of how it will impact or improve the regulated safety features when installed.

Comma One Front

11. Describe in detail how the comma one impacts a vehicle’s rearview mirror, including whether it requires removal of the rearview mirror or the extent to which it block or obstructs the rearview mirror.

12. State your position on how the comma one does or does not affect a vehicle’s compliance with FMVSS No. 111, Rearview mirrors (49 C.F.R. 571.111), and provide any supporting information or documentation to support your position.

The rearview-mirror replacement appeared to be illegal and NHTSA thought the same based on the questions above. It’s possible Hotz would be able to get by with using the screen as the mirror, but we’ll never know since he cut off communication with NHTSA immediately.

13. State the date on which you currently plan to begin selling the comma one, and provide a list of all retailers and/or website through which you anticipate selling the comma one.

Hotz already announced he would sell before the end of the year, so all he needed to do here was provide an exact date to NHTSA.

14. State the date on which you currently plan to begin shipping the comma one.

This is also a simple question and could be based on the date above depending on how long it would take to produce the initial batch of devices.

15. Provide any other information which you believe supports the safety of the comma one.

The last questions allows Hotz an open-ended platform to provide as much supporting documentation as he would like to prove his case and does not limit him in any fashion. The fines set in the request also clearly state that they will only be enacted if a response is not received by November 10th. There is nothing in the letter that states he must cancel the product or even have a full plan to be compliant by that date. This special order appears only to be a request to start the compliance process with NHTSA.

Based on how simple the questions are, and the knowledge I have even as an outsider, I believe the request could have been completed in a day. Instead, Hotz chose to let his ego show again, posted the document publicly, and cancelled the product.

He had lots of great ideas that could’ve been beneficial to the industry. Maybe he’ll share some of them when he decides to join the grown-ups and fill out a simple information request.

Get the latest TTAC e-Newsletter!

Recommended

89 Comments on “These Are The 15 Questions That Caused George Hotz to Cancel Comma.ai...”


  • avatar
    Kenmore

    You don’t often see oxygen tubing juxtaposed with dark hair.

  • avatar
    JimZ

    “Hotz should be able to answer this easily as his hand-off procedure was recently praised by testers who’ve been inside the test vehicle.”

    The only bit I heard was Alex Roy saying it was “better than everyone else’s.” My response was “based on what?

    “This is important since Hotz was planning to use an OBD II device that would plug into supported vehicles. Since this is a universal standard, the device could be plugged into any other vehicle and could cause unintended consequences since Hotz was forcing messages over the CAN-BUS.”

    it’s important to remember that CAN is a hardware and *transport layer* standard. It basically lays out “this is how you send/recieve a message, these are the speeds you can send/receive messages, and this is the hardware you need to do it.” The critical thing is that the set of standards for CAN does not specify what those messages must be. OBD II diagnostics for emissions-related things are the only rigidly standardized things, and I don’t believe the Comma.ai thing can do what it was meant to do strictly via OBD communication. most cars have more than one CAN network (chassis & body,) and manufacturers are free to do whatever they want on those networks. Hotz would have needed access to the other networks and plugging one meant for Honda into e.g. a Ford would likely corrupt the bus and shut parts of the car down.

    • 0 avatar
      DevilsRotary86

      “it’s important to remember that CAN is a hardware and *transport layer* standard. It basically lays out “this is how you send/recieve a message, these are the speeds you can send/receive messages, and this is the hardware you need to do it.””

      Slightly off topic, but this just made me think of the few times that I look at CAN bus. Whenever I look at CAN bus it always reminds me a little of RS-485. Mostly in how both are multi-drop off of one TX/RX pair and both require terminating resistors.

      Although the only time I actually see CAN bus is when I go to embedded computing conferences. My industry almost exclusively uses RS-485 + Modbus, or just old fashioned RS232.

      • 0 avatar
        JimZ

        it wouldn’t surprise me if that’s what inspired the hardware side of CAN. Differential signaling, 120 ohm termination, etc.

        • 0 avatar
          xander18

          CAN is designed to be crazy noise immune. At every layer of the standard the primary purpose is on message integrity. Differential signalling is great for this because the units are actually looking at the difference between the two lines. If electrical noise is injected into the signal it will bias both lines the same and the difference between the two will be relatively unaffected. The terminating resistors cut down on high frequency reflections.

          It’s a pretty badass protocol. Hampered by low speed but compared to something like Ethernet or wifi it just doesn’t need that much bandwidth. CAN-FD is poised to improve that as well.

          • 0 avatar
            DevilsRotary86

            Considering I work with devices that use 9,600 baud on RS232, I am pretty familiar with very low bandwidth applications. Don’t need much bandwidth when you are tossing around messages that are only 4 to 30 bytes long.

    • 0 avatar
      DevilsRotary86

      “The only bit I heard was Alex Roy saying it was “better than everyone else’s.” My response was “based on what?”

      I just read something on TheDrive.com from Alex Roy himself, and your comment came to mind when I read this tidbit:

      “Hotz’s initial tweets suggested his move was in response to an inquiry from NHTSA, stating that he “would much rather spend life building amazing tech than dealing with regulators and lawyers,” and that Comma.ai would be “exploring other products and markets.” Twenty-four hours after the cancellation, Hotz called me from China. Asked if he was genuinely intimidated by NHTSA’s letter, he said, “I’ve got two words for you: stealth mode.”

      “Stealth mode” sure doesn’t sound like he’s backing down.”

      hoo boy, this could get interesting.

  • avatar
    LS1Fan

    “Instead, Hotz chose to let his ego show again, posted the document publicly, and cancelled the product.”

    Because there was no product in the first place.

  • avatar
    mike1dog

    There’s a big difference between an electronic device, that if it fails does nothing more than make the owner of the device mad, and a device that could affect a thirty-five hundred pound vehicle going seventy miles an hour. These questions seem reasonable. I’m guessing Tesla got a similar letter about autopilot?

  • avatar
    Pch101

    “it’s apparent the questions are reasonable and easy to answer.”

    The questions are reasonable. But easy? No. For example, you would not want to answer the FMVSS compliance questions without running them past counsel and carefully phrasing your answers.

    What is likely is that the device hasn’t really been tested. He simply hasn’t had the time, budget or staffing to make sure that it really works, and he probably never will. Hotz likes to claim that this stuff is easy, but it isn’t.

    • 0 avatar
      JimZ

      building a complex system seems to follow an 80/20 rule. 80% of the work (getting it to function) is “easy.” The remaining 20% (getting it to work reliably, consistently, safely, etc.) is *damned hard.*

      Guys like Hotz get to about 70-80% then throw hissy fits and give up when told they need to do the last 20% too.

      • 0 avatar
        FormerFF

        Actually, I think the 90/90 rule applies: The first 90 percent of the project takes 90 percent of the time, and then the last 10 percent of the project takes up another 90 percent of the time.

      • 0 avatar

        Yeah that sounds about right. I have found that too working for small manufacturing companies. Developing a working product not too bad, developing a working product that doesn’t fail under 99.9% of the conditions it will encounter lot harder.

      • 0 avatar
        slow_poke

        for me it always seems that getting 80% of the functionality takes 20% of the time. let’s call that a “demo.” and the rest of the functionality, edge cases, reliability, etc… what JimZ said, takes the other 80% of the time.

        i feel we see that time and time again in the tech industry. “i’ve just about got it…”

    • 0 avatar
      sirwired

      Well, the questions should have been easy to answer, had he given any thought at all to regulatory compliance.

      I think the one asking about review mirror standards really sums things up; it’s a detailed and clear section of the standards that he obviously must demonstrate compliance with, given that he’s replacing the factory mirror. If he was planning on FMVSS compliance at all, he should have had someone going down the relevant parts of that section like a checklist before even creating a prototype of the production version.

    • 0 avatar

      Just remember folks, one has to be an adept, a member of a special guild made up of our intellectual betters, in order to be able to understand the law. Also, there’s no rent-seeking in how the legal industry uses the writing of legislation and how it controls the courts and regulatory agencies to create perpetual employment for its members.

      • 0 avatar
        Pch101

        Sulk less, learn more.

      • 0 avatar
        JimZ

        “Just remember folks, one has to be an adept, a member of a special guild made up of our intellectual betters, in order to be able to understand the law.”

        yes, we call those people lawyers. they get paid to do that.

        there’s also another special guild, who are trained in the ability to understand and use scientific principles to design/discover solutions to real-world problems. and they have to understand how to work within the constraints of that real world. We’re called engineers, we get paid to do that. We’re also paid to make sure the stuff we design will work properly and safely in its intended role. And there are some of us (not me) who will put their careers and freedom on the line to evaluate and sign off on these things (Professional Engineer certifications.)

        George Hotz is none of that. he’s a tinkerer who thinks he got something to work, and he was ready to throw it out into the world without even knowing how well it would work in the real world.

        normally I like the stuff you write, Ronnie, but I can’t stomach you trying to turn this into a condemnation of those “EEEEEEVIL REGULATIONS.”

        • 0 avatar
          Pch101

          He’s not complaining about lawyers. He’s whining because his opinions that are based upon hot air and politically-charged nonsense have no value.

          He believes that his own views aren’t given enough respect. The fact that his positions are consistently false makes no difference because his little feelings are hurt. (That chip on his shoulder must hinder his ability to think clearly or research these subjects.)

      • 0 avatar
        sirwired

        Really Ronnie? Bozi goes over the exact questions in detail, and shows how they are not, in fact, legal mumbo-jumbo, and the best you can do is a generic rant against lawyers?

    • 0 avatar
      psarhjinian

      This was designed using the normal SV “move fast, break things, get funding, stall regulators” fashion that works quite well for loosely-regulated industries.

      It’s what you do when you want to get an product to market and what matters is getting it done quickly so that you can maximize what funding you have and get the needed PR for your next round. Uber has done this to great effect: moving in, moving fast, doing things 80% right and engaging in stalling actions with regulators while they develop the succeeding and/or complimentary products.

      The problem is that this is a much more regulated industry because, unlike apps, ridesharing, food delivery or virtual hotels, this kind of technology could result in lots of people being maimed or killed. We don’t let companies put drugs on the market, or build nuclear reactors, without sufficient testing for this exact same reason.

      It’s the same issue that Theranos faced, and Theranos got much further than they really should have: they were selling a service that didn’t really work, in hopes they could raise money to make it work, or get into something else by the time this particular house of cards toppled. What’s galling is that they didn’t really seem to care that fake blood test results might result in people dying, just as geohotz doesn’t seem to realize that ,ai might run someone down.

      It’s a good thing that regulators slapped this one down before it got too far. I certainly wouldn’t want to see the Theranos of self-driving cars wreck an entire industry just because some sociopathic libertarian man-child thought that beta-testing this on public roads was just fine.

  • avatar
    sirwired

    I think the fact that regulators had the temerity to ask any questions at all is what led him to cancel. I mean, HOW DARE the Feds try to verify that something that can take over braking and steering not kill drivers, passengers, and innocent bystanders! The nerve!

    As you pointed, out, (both here and in the original article), these are not questions that should have been at all difficult to answer if the product was, in fact, going to ship by the end of the year as promised, had he given any thought at all to regulatory compliance.

    I thought a couple of the questions neatly summed up what the Feds were thinking:

    “10. Have you done any analysis or testing of the impact or potential impact of comma one on the vehicle’s compliance with FMVSS? If yes, please describe the analysis or testing in detail and provide supporting documentation. If no, describe why not.”

    Translation: We have this thick book of safety standards for vehicle safety that anybody can download called the FMVSS. Perhaps you’ve heard of it? It’s kind of, like, our whole job as an agency. If you’ve heard of it, did you bother to read it? It’s kind of important. Obviously a lot of it doesn’t apply to you (we know your device does not effect, for instance, the tires or windshield wipers), but there are a lot of parts that certainly are relevant to your device.

    “12. State your position on how the comma one does or does not affect a vehicle’s compliance with FMVSS No. 111, Rearview mirrors (49 C.F.R. 571.111), and provide any supporting information or documentation to support your position.”

    Translation: That thick book we were just talking about? The Table of Contents has a section labeled “Rearview Mirrors”. We know the FMVSS is a pretty large and complicated document, but that’d be a good place to start for a device that replaces the standard mirror. If you don’t have detailed answers ready to go quickly (indicating you’ve already done the specific tests the standard requires), our confidence level in the idea that you’ve read ANY of it is pretty darn low.

  • avatar
    Car Guy

    The government doesn’t have a clue how to regulate this stuff so they throw out a “voluntary” guideline (which we all know isn’t really voluntary) and ask companies to “prove” their systems are safe. How do you prove something without a objective standard to measure against? Isn’t that their job to set the standard like every other safety system?

    No matter what you tell them, they can arbitrarily deny the request if there is something they don’t like. The whole thing stinks and does nothing except stifle innovation.

    • 0 avatar
      Kenmore

      “The whole thing stinks and does nothing except stifle innovation.”

      Time for me to watch The Jerk again. Love Steve Martin!

      • 0 avatar
        Car Guy

        You are great at trying to run people down but avoiding the question at hand. SO I’ll ask again because you seem slow. How would you prove something is safe if they failed to give a standard to measure against?

        • 0 avatar
          Pch101

          Among other things, NHTSA just released a policy document that is 116 pages long on this subject.

          Of course, we have the FMVSS.

          All of this stuff is easily found on the internet. So how is it that you didn’t know about any of it?

          • 0 avatar
            Car Guy

            @pch101- show me a FMVSS standard that dictates the performance of automated vehicles. A government “guideline” is not a FMVSS.

          • 0 avatar
            Pch101

            Er, the FMVSS applies to all vehicles, including the autonomous ones.

            Per a guy who represents the autonomous car lobby: “(FMVSS) standards include 286 references to human drivers and mandate traditional design features such as steering wheels, displays and pedals. Any fully self-driving vehicle would violate more than one-third of those standards, not to mention half of the Series 100 Crash Avoidance standards built into the compliance structure.”

            http://thehill.com/blogs/congress-blog/technology/299286-innovators-must-speak-out-on-us-dot-autonomous-vehicle-policy

            Stick to engineering, and leave the legal discussions to someone else.

          • 0 avatar
            Car Guy

            @Pch101 – The vehicle would still meet all the FMVSS standards on the books and the system this guy invented would be able to supplement the driving task with the driver still in the vehicle. No FMVSS are being violated. For what the system does, there is no FMVSS to dictate performance. That’s where I see the problem. There is no regulation yet the government is using a threat of a $21,000 daily fine to discourage this company from releasing their product. Show me something that says your radar must detect an oncoming vehicle at “X” meters away. Or your camera must discriminate between a passenger car and heavy truck “Y” times a second. That is objective and reasonable. The government “guidline” offers nothing objective.

          • 0 avatar
            Pch101

            It’s clear that you don’t understand the issue. And I’m pretty sure that you never will.

          • 0 avatar
            Car Guy

            I understand the issue well. The quote you pulled is related to fully automated vehicles that do not have a traditional steering wheel, brake pedal, etc. Yes, I agree that’s a problem and under the current rule set would be non-compliant. But the system under discussion here takes a fully compliant base vehicle and provides a driving supplement. There are no performance rules for this kind of system and there are currently no laws prohibiting them. A government “guideline” with no performance standard is not enforceable but they are using threats of fines to enforce something they currently have no regulation covering. That’s the whole point here.

          • 0 avatar

            It modifies systems that fell under the rules that would make it fall under the rules as well.

          • 0 avatar
            Pch101

            “I understand the issue well.”

            I’m sure that you would like to think so, but some of us can see that you don’t.

            NHTSA has authority over all vehicles, not just the autonomous ones. But I can see that you have an either-or mindset and that you do not comprehend that having a set of voluntary guidelines does not negate the existence of those regulations.

          • 0 avatar
            Car Guy

            At the end of the day only safety standards that have an official final rule published are enforceable. The government is trying to short cut a well established process and using “guidelines” and threats of fines to intimidate people into getting their way without doing all the due diligence they are required to preform. That’s what is going on here and it’s wrong. I’m not against rules that make sense and have been fully vetted. This guideline has not been.

          • 0 avatar
            Pch101

            Comma.ai has been asked to explain how it is complying with the FMVSS generally and FMVSS 111 specifically, and Hotz won’t do it. Perhaps he can’t do it.

            You can keep repeating yourself, but being redundant only makes you tedious, not accurate.

        • 0 avatar
          Kenmore

          “trying to run people down but avoiding the question at hand”

          Well, I am being redundant in running *you* down but otherwise I’m just staying out of the way while the pros here solo.

    • 0 avatar
      sirwired

      The FMVSS aren’t voluntary. Where did you get that idea?

      Yep, the NHTSA hasn’t fully baked standards for autonomous systems yet, but there are plenty of existing regulations that totally apply here.

      For instance, take the standard on rearview mirrors they were asking about: That has a nice list of clear, objective, criteria that a rearview mirror (or substitute system) must meet. Proving compliance is rather straightforward.

      Of course, there are a bunch of other standards beyond that one that he’d need to comply with; exactly what you’d expect for something that can take over braking and steering.

      • 0 avatar
        JimZ

        he’s just another “regulations are evil” bot. I’m surprised he didn’t work in “disruption” and “job creators” also.

        • 0 avatar
          Car Guy

          No, I’m an engineer who works with standards and data. The burden is on the government to prove something isn’t safe – not the other way around. And given there is no FMVSS related to this inventors system, he was given an impossible task and no way he could answer in their unrealistic time frame or risk many thousands in fines.

          • 0 avatar
            sirwired

            “And given there is no FMVSS related to this inventors system, he was given an impossible task and no way he could answer in their unrealistic time frame or risk many thousands in fines.”

            “No” FMVSS related to his system? There are plenty of standards relating to steering and braking (systems which talk to this gadget), and they even explicitly pointed out the rearview mirror standard he needs to comply with. (I think that was the real “test” question: it’s a straightforward “checklist”-style regulation and if he couldn’t answer that quickly, it’s strong evidence he gave no thought whatsoever to regulatory compliance.)

            Which one of the DOT’s questions was “impossible” or did they give an “unrealistic time frame”? These are pretty straightforward questions he should already have a good idea how to answer if he really planned on shipping by the end of the year.

          • 0 avatar
            anomaly149

            I hope you’re not an engineer! The onus is on engineers to demonstrate safety during development, not on everyone else to find out the hard way later! (Or the government to prove that it’s dangerous)

            This attitude wouldn’t fly for a minute in a real safety organization, hence the product is withdrawn.

          • 0 avatar
            JimZ

            No, that’s not what’s in play here. He was preparing to sell a device to give existing (FMVSS certified) human-driven cars some level of autonomous capability. They stepped in to say “Hey, you need to tell us how this works before you can sell it.”

            in case you hadn’t noticed, nobody else is SELLING autonomous vehicles yet, which is why they’re still in the “guideline” stage. Hotz was about to sell something, and NHTSA wanted evidence his device wasn’t going to make cars equipped with it operate in a less-safe manner.

            you can bet your *ss no automaker is going to actually sell an autonomous vehicle without providing the same data to NHTSA’s satisfaction. Even Tesla’s recent announcements are still bluster since they won’t actually be enabled for some time.

          • 0 avatar
            Coopdeville

            From earlier: “A government “guideline” with no performance standard is not enforceable but they are using threats of fines to enforce something they currently have no regulation covering.”

            I don’t know how the NHTSA works but I perform regulatory compliance in a field governed by the IRS and on the periphery the DOL, and I can tell you with confidence that enforcement in my industry often hinges on guidelines as opposed to formal regulation. The IRS can and does change the rules (or expand them) verbally at conferences and in written opinions. If you’re lucky they then back that up with a Revenue Procedure which codifies the “guidance.” Woe unto the audited employer or industry pro who does not conform to “mere guidelines.”

          • 0 avatar
            NickS

            @carguy. Your position is in stark contrast to that of every engineer I have met (I direct engineering teams btw). Usually business people make the argument that their product should be assumed to be safe unless someone else proves it unsafe. Engineers are usually the one who get pressured to overlook or violate their professional ethics and/or safety concerns to meet a business goal.

            There are some “engineers” who think their prototype is all it takes to start selling something, but we don’t hire those.

            Generally speaking, anyone asking questions for any brilliant idea is potentially stifling innovation but in reality more often than not they are essentially exposing the innovation as a solution that only solves a very narrow and conveniently chosen set of constraints, but plays in a much bigger problem space.

            The model you subscribe to is the reason we have so many unregulated chemicals. Do you really want to be the one who provides the proof that one of them killed one of your family members? Or would you rather have the chemical company at least show that they thought through the safety issues of a new chemical they are about to bring to market?

    • 0 avatar
      turbo_awd

      does nothing except stifle innovation.

      Nothing except stifle innovation? Nothing else at all? How about saving millions of lives, etc?

      You just scream “shill”..

      • 0 avatar
        bikegoesbaa

        “The burden is on the government to prove something isn’t safe – not the other way around.”

        Here in the real world it is definitely the OEM’s responsibility to demonstrate safety and compliance with applicable standards.

        Safety standards are sometimes written qualitatively. This is a good thing, as it gives more latitude to the designers to come up with new and better solutions as long as they can make a sound case for these solutions being safe.

        It also discourages manufacturers from overlooking problems with designs that meet the letter of the standard but not the intent.

    • 0 avatar
      Car Guy

      All this proves is the government can bully small inventors out of the market with the threat of fines if you give answers they don’t like. Would you seriously risk that if you had a fledgling company with little resources? Perhaps if you have the bankroll Elon Musk has for Autopilot you can afford to put up a fight. For the little guy there is no chance……

      • 0 avatar
        bikegoesbaa

        The problem isn’t that they didn’t like his answers, it’s that he didn’t bother to give *any* answers and instead left in a huff.

        If you intend to sell an accessory that will control a car on public roads you need to convince the mean old gubbamint that it’s not going to kill an abnormal number of people. I bet you are in a small minority who does not regard this as a good thing.

        Many small technical entrepreneurs are successful in the current regulatory environment. They are successful because they educate themselves about the applicable regulations and address them during their design process like they would any other technical challenge.

        Otherwise, don’t start selling auto-pilot systems for airplanes and act shocked when the FAA shows up asking questions.

      • 0 avatar
        mcs

        @carguy:

        There is a path for small companies that will keep you off of the government’s (and competitors patent attorneys) radar. Develop your product, but concentrate on building a patent portfolio rather than bringing it to actual production. Then license small pieces to the big guys. Find a niche and focus on it. That’s been my approach and I think it’s the way to go for small organizations without a multi-billion dollar budget.

        Look at Brembo. They aren’t building entire cars. Sure, they have to deal with government brake standards, but they can focus on that one type of product and do it better than anyone else.

        Battery labs are another example. They are building new technology batteries in limited pilot production, but for mass production, they are licensing to the big companies.

        There are plenty of small niches to focus on in autonomous vehicle development. Concentrate on a few small areas, patent your work, then license the pieces/code libraries. You can also keep your costs much lower and sometimes even avoid the need for outside investment.

  • avatar
    PrincipalDan

    Before I refuse to take your questions, I have an opening statement. – Ronald Reagan

  • avatar
    turbo_awd

    Hotz thought he could go at this like Uber and AirBnB – don’t even bother reading up on regulations, and ignore any that are brought to your attention.. If you bullshit your way long enough, eventually people will have it in their cars and like it, and then you can force the hand of regulators.

    Sorry GeoHot – the internet/video game playground (ps3, iphone) doesn’t always translate nicely to real life.. (people’s lives in your hands). If you thought Sony was being a dick about hacking the ps3, just wait until someone sues you for causing their relatives to die..

    • 0 avatar
      DevilsRotary86

      “Hotz thought he could go at this like Uber and AirBnB – don’t even bother reading up on regulations, and ignore any that are brought to your attention..”

      In general you are right, but Uber was right about cab regulations being ridiculous and unfair. For my little corner of the USA, Dallas, Uber’s fight to set up shop in the market ended up ushering in meaningful reform to for-hire laws.

      Before the regulations basically said (without actually saying it) ‘want to be a taxi driver? then sign up for employment with one of Dallas’s 3 friendly taxi companies!’ Now that has changed so that anyone (Uber included) is welcome to hang a shingle as a taxi company so long as they pay the city a yearly $300 fee, have a background check at the company/individual’s expense, have an inspected car, and have commercial insurance.

      It was a bit of drama to get here, including indirectly leading to the resignation of the city manager, but I think it’s worth it. This is of course the experience of just one city and I know it varied from city to city.

      • 0 avatar
        turbo_awd

        DevilsRotary – Uber may have helped in your case to improve SOME things, but they’re still doing their best to shaft all their drivers AND replace them with autonomous cars.. I.e. using the profits from the current drivers, not to improve the company, but to put them out of work. To me, making some improvements, but shafting the people who helped to get you there, as soon as it’s possible, isn’t a company I want to do business with. Not to mention the constant cutbacks of driver wages.

        Edit: cut out a rant that was a little too over-the-top. I feel better now, I really do :-)

        • 0 avatar
          bikegoesbaa

          To be fair, developing ways to deliver the same service for lower costs *is* improving the company.

          There’s not a company in the world that is not interested in ways to get more work out of fewer employees. Most of them are pursuing automation to various degrees in order to bring this about.

          Uber is just more upfront about it.

        • 0 avatar
          Paul Alexander

          Uber’s an overcapitalized cab company that calls itself an ‘app’. It’s main purpsose, though, and the reason it’s been able to raise so much money, is as the Trojan Horse for the ‘disruption’ of existing employment laws.

  • avatar

    I work in medical devices, and this is pretty much the same list of things we deal with from the FDA.

    “Show us how it’s designed to work and in what situations; show us the testing that proves it works the way it’s supposed to; tell us what foreseeable misuse you’ve accounted for.”

    Occasionally they miss the point or get hung up on some irrelevant detail, but it’s really not that hard if you document and fully vet your design.

    While “could be answered in a day” is certainly an overstatement, a couple weeks to put together all the documentation wouldn’t be all that bad.

  • avatar
    tylanner

    Question 4 is extraordinarily hard to comprehensively answer…and deliver on.

    To even develop the nomenclature and vocabulary necessary to accurately describe and defend the intended bounds of a self-driving device is a monumental task given the endless variability and inconsistent nature of our roads and highways.

    You could theoretically keep it simple in your answer but the more general you are in your definitions the more complex and smart your device must be to process the variable inputs into a general, solvable case that results in a safe driving path.

    A simple device such as comma would need to be very, VERY demonstrative in answering these questions which, I think, would expose the basic flaws and impracticality of this cheap and simple answer to a very difficult problem.

    Hotz got off the elevator on the ground floor…these are the most fundamental design considerations that any regulated product, with grave public safety implications, must work through.

    • 0 avatar
      sirwired

      Question 4 certainly does capture the difficulty of making such a device. As you said, if they’d put any thought at all into proper design of the product, it’s pretty straightforward to answer (even if the answer might not be simple), but if you are trying to throw a product out the door the quality of a typical smartphone app, it would, of course, be impossible to answer.

  • avatar
    5280thinair

    For Hotz to just cancel the product outright this close to shipping to customers in itself suggests something fishy going on. Product manufacturing involves contracted purchase and lead-time agreements, and if he was truly that close to volume production he should have substantial money on the line with suppliers and assemblers. That’s a powerful disincentive to bail.

    Either he bet the farm on not having regulatory scrutiny applied to the device, or he was at best prepped for small-scale production (e.g. hand assembly) where the financial penalty for walking away isn’t too severe.

  • avatar
    05lgt

    Who’s money did he steal before using this as an excuse to cut and run? the only other answer I can make fit is that he never previously ran any of it past competent legal representation and when informed that if he killed people while ignoring regulations he could spend time in FITB prison… he ran.

  • avatar
    PeriSoft

    The idea that a single camera can drive a car on anything approaching current tech is so absurd it beggars belief that people ever gave this guy any credence at all.

    Even the best current systems require LIDAR and multiple cameras, and still fail when they can’t see road markings.

    I was driving in town last night in the rain. It was essentially impossible to see any road characteristics at all; any visual cues were *entirely* subsumed by reflections and glare. The only way I could drive was by knowing where I was vs buildings I was familiar with, and context awareness – subconscious understanding of traffic flow, rules of the road, and the way that roads in my area are laid out. I’ve been in different areas of the state or country where slightly different conventions in traffic flow meant that bad visibility made it extremely difficult to drive at all. Now imagine being a robot with bad vision and everything is new all the time!

    Even a high end SLR would have produced images completely useless for navigation in that situation. And it wasn’t even the worst. Claiming that you’re going to do level four autonomy with a cell phone camera is farcical, and anyone entertaining the notion that it was ever remotely possible is complicit in what was essentially a scam from the start.

    • 0 avatar

      I’d be interested to hear what the folks at Mobileye genuinely think about the Comma device. They’re an important vendor of driver assist systems today but in their early days they had a bit of a hard time convincing OEMs that a camera/software system could work as well as radar and LIDAR.

      • 0 avatar
        PeriSoft

        A multiple visual camera system might be coaxed into generating an OK terrain map. But the real problem is that terrain mapping isn’t even the biggest problem; there are problems upon problems and each level is not only in the “Really Freakin’ Hard” category, but also the “Nobody Is Even Trying Yet” category. How to handle humans directing traffic by hand? How to describe where to park a car in an empty field? How to handle situations where traffic flow explicitly counters lane markings at different times? How to handle extremely poor visibility (there are plenty of situations where humans drive essentially based on memory and, as I said, context, rather than on realtime input)?

        Sure, if you still have a steering wheel and pedals you can just hand control back to drivers. But when that control hand-off is in realtime, evidence suggests that the spike in danger goes through the roof. And, taken to its conclusion, this approach would result in a nation of drivers who only control their cars when conditions are poor: Not the best way to reduce accidents.

        I can envision a future where cars will chirp at you as you enter big, well-maintained traffic systems, and operate autonomously until they’re on surface streets. But anything beyond that is pretty much either all or nothing, and ‘all’ at this point is a very, very, *very* tall order.

  • avatar
    fishiftstick

    There’s a bigger picture.
    Say an autonomous car requires intervention only once per million miles. Human nature says the driver won’t be paying attention, so the car will crash. But only once per million miles, so that’s okay, right?
    Yet the average car does over 100k miles in its lifetime. At that rate, more than 1 in 10 autonomous cars will crash due to failure of autonomous driving technology.
    Compare that to Takata. They recalled 41 million cars because a few hundred airbags exploded. To get to that failure rate, an autonomous system would have to require intervention just once in a trillion miles.* To reiterate, this is the failure rate where all your cars get recalled and you go bankrupt.
    On the other hand, human drivers err much more often than that, so an autonomous system as faulty as Takata airbags would be a massive improvement that would prevent injuries and save lives.
    So here’s the question: Is it reasonable to regulate autonomous cars to the same standards as other auto parts, or does the regulatory regime require a complete overhaul?

    * Takata = about 1 airbag explosion per 100,000 cars; 100,000 cars x 100,000 miles = 1 trillion miles per defect.

    • 0 avatar
      bikegoesbaa

      I think it’s reasonable to hold them to a much lower standard than simpler single-function parts like air bags.

      I don’t require autonomous vehicles to be perfect, or near perfect. I just want them to be be better on average than humans. Humans, as it turns out, are terrible drivers. In the US we kill ~30,000 people every year.

      As soon as the OEMs prove that autonomous vehicles can reduce the death rate (and lesser crashes and injuries, of course) below the current average meatbag standard I’m all-in on the technology.

    • 0 avatar
      JimZ

      apples to oranges. Not every Takata airbag will deploy. in fact, I’d say a fairly small fraction of them will be called to deploy. The reason they’re all recalled is because 1) you can’t predict which airbags will need to deploy, and 2) Takata has no way of determining which airbag inflators are more/less likely to rupture apart from mumbling a bit about humidity. so with that kind of uncertainty, NHTSA has to assume they’re all dangerous.

      the more interesting (and telling) statistic is “what percentage of *deployed* Takata airbag inflators failed/ruptured?”

    • 0 avatar
      NickS

      @fishif
      It is not obvious to me how that makes a device worth deploying. You could use the stats for or against a product. The real question is what is the threshold for safety. This guy seems to have done zero thinking on that front.

      Forget about people in the abstract. Do you want to be behind that wheel when the airbag deploys, or when that device corrupts the CAN bus and one of the stock safety systems fails to work? Or when that device gets confused and causes a cascade of failures that leads to your BFF getting mowed down?

      The problem with the once-in-a-million-miles argument is that it could happen on mile 23, or on mile 230,000. You are hoping it is the latter but you really don’t know.

      • 0 avatar
        bikegoesbaa

        “Do you want to be behind that wheel when the airbag deploys, or when that device corrupts the CAN bus and one of the stock safety systems fails to work?”

        No, but I also don’t want to be in a car when the human driver crashes it due to inattention, distraction, or fatigue. Or when it gets hit by a drunken human driver who crosses the median.

        • 0 avatar
          NickS

          Yes, but this device is nowhere near a solution that will shave off all the fatalities caused by human error. If anything, permitting a system like that to be sold without asking anything about its safety priorities is akin to never testing humans for their driving skills, at least once.

          What you may be thinking is a goal far down the road where we have capable autonomous systems, but this little box is simply not it. It may end up making those same flawed human beings (that I also don’t want Bing behind that wheel) even more dangerous.

          • 0 avatar
            bikegoesbaa

            I agree, this device is at best woefully misguided and at worst an outright scam.

            But I am willing to accept automated cars in general with a failure rate much much lower than the “one fault per million miles” proposed in the first comment in this thread.

          • 0 avatar
            JimZ

            nobody’s using this to damn autonomous cars. we’re just saying it ain’t going to come from some on-the-spectrum geek who thinks his s#it doesn’t stink.

      • 0 avatar
        28-Cars-Later

        “Do you want to be behind that wheel when the airbag deploys”

        That depends, did the car hit something or was there a sensor malfunction?

        • 0 avatar
          NickS

          I was referring to a defective Takata airbags.

          I don’t know how others think about this but I obviously want airbags and willing to accept the risks that come with one that functions as designed. Am I willing to drive a car with an airbag that may or may not function as designed? True, what are the chances, but does it really matter if the next one is you or one of your loved ones.

      • 0 avatar
        fishiftstick

        Stats are how I measure costs, benefits and relative safety. How would you do it?

        I don’t want my BFF killed in a plane crash either. But if he asks me whether it’s safer to drive or fly, I’ll tell him flying is safer, because statistically it is.

        And just to be clear, I am not defending this device. The only regulatory regime its maker could handle was none, and that will never be how these things work.

  • avatar
    Jimal

    Never, ever, question a special snowflake.

  • avatar
    jdmcomp

    This man has no business being in business, he is an idiot if he thinks he can win at this war. I do believe that this is the most stupid move of the year.

Read all comments

Back to TopLeave a Reply

You must be logged in to post a comment.

Recent Comments

  • ajla: You could do that on several FWD GM cars of the late 80s to early 90s as well.
  • MRF 95 T-Bird: I owned a 80 Toronado Diesel in Grey Firemist with the claret velour interior. The seats were quite...
  • Scoutdude: There are two different things going on here. The recall is for the X and panels that weren’t...
  • ToolGuy: redgolf, thank you for that link. Some eye-opening quotes there. “We want to lead in this space. We don’t...
  • Lightspeed: Maximum Brougham! Nice car, though I prefer the more nimble B-bodies of the era. The whole Chev engine in...

New Car Research

Get a Free Dealer Quote

Who We Are

  • Matthew Guy
  • Timothy Cain
  • Adam Tonge
  • Bozi Tatarevic
  • Chris Tonn
  • Corey Lewis
  • Mark Baruth
  • Ronnie Schreiber