Slashdot Mirror
Comma.ai Shelves Self-Driving Device After Regulatory Warning (reuters.com)
Comma.ai founder George Hotz, who has spent the good part of his past year criticizing competitors and their technologies, sent out a series of tweets Friday, saying that Comma.ai, a startup that aimed at offering semi-autonomous driving system, will be pulling out of the U.S. market in response to requests from federal regulators. From a Reuters report: The intervention, by the U.S. National Highway Traffic Safety Administration, came before Comma.ai began marketing its device. It is the latest signal that regulators want more control over the development and deployment of self-driving vehicle systems by vehicle manufacturers and suppliers, after a period in which they took a largely hands-off approach. The NHTSA on Friday disclosed an Oct. 27 letter to Comma.ai stating that the agency is investigating whether the company's device, called Comma One, complies with federal regulations. The letter and an accompanying special order demanded that Comma.ai provide the agency with information about the device and warned that the agency could prohibit the sale of the system if it were found to be defective.
Beats having to make it work... I guess.
Company and product name sounded suspiciously similar to coma...
ELOI, ELOI, LAMA SABACHTHANI!?
The letter and an accompanying special order demanded that Comma.ai provide the agency with information about the device and warned that the agency could prohibit the sale of the system if it were found to be defective.
So this "intervention" was simply a letter asking for information and a warning that the NHTSA could order the developer to pull their product if found to be defective -- that's it. So, they preemptively pulled their product from the US market. I can only assume the product was either vaporware or defective.
It must have been something you assimilated. . . .
Sometimes it absolutely stuns me how large the differences are between 'business leaders' and myself. I would have assumed from the start that the government would have to act in a way similar to what the letter specifies. In contrast, along comes George Holtz who didn't see this coming at all, and it comes at such a shock that he feels he needs to shelve his business? Seems very out of touch with reality and a functioning society to me, unless of course there is something else to the story.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Well I see (s)he fixed it now. I was sort of hoping I hadn't kept up and "a service of tweets" really was a freshly coined collective noun. Editor incompetence is a much less satisfying answer. But then I'm a fan of collective nouns, I think we should be actively inventing them.
Disclaimer: This is my personal opinion. I am not authorized to speak for my company.
Most people have no idea what "information" is required when approval is required for a life critical system involving software.
A "Stall Warning Card" with a 4k ROM program required the production and submission of more than 40,000 pages of documentation. It required more than 100 times more effort to produce the paper than the software.
I suspect that once Comma.ai looked at the regulation, the standards reference and calculated the volume of documentation required to prove compliance, they decided that it was too expensive to do business in America.
This was an unfinished and unreleased product. I think they were out of line. Enforcing that products are safe is a good thing, and I don't have a problem with the NHTSA doing that. But this level of fine (up to just over 100 million USD) seems ridiculous for a product that hasn't been completed / released. Essentially they are fining someone with an idea they are trying to explore / develop, even though that person is not selling anything yet, and may never sell anything.
Even the questions they asked seemed crazy at this point. They wanted him to describe how the user interacts with the system. The system isn't complete! It hasn't been finished. It may not even have an end user UI yet. Don't know what valuable information they thought they were going to get here. But they start by fining him $20,000 EVERY DAY! Come on. That is over the top and ridiculous.
Yes! Let's start adding to the autolog of collective nouns!
this signature has been removed due to a DMCA takedown notice
NO NO NO!!!
Sorry but it HAS to be:
"a fleet of tweets".
As in The candidate last night launched a fleet of tweets to discredit their opponent...
You have the right to remain sentient. If you give up the right to remain sentient, you will be elected to public office
Twit of tweets?
Tweet of twits?
Faster! Faster! Faster would be better!
My vote is for "a tweet of twits", sounds like a fleet of ships.
Pain is merely failure leaving the body
Whew. Good thing you had that disclaimer. I thought for a second you were speaking for your company! Why do people bother writing garbage like this? No one knows who you are or what your "company" is.
Take these "self driving" efforts for what they are: a PR exercise to swindle clueless investors. It is true that 90% of the problem of autonomous driving is solved. It's also true that the remaining 10% are exponentially hard and won't be solved in the next 20 years. By 10% I mean such trivial (to a human) things like driving in bad weather, or on dirty pavement, or figuring out if the object on the road is a plastic bag or a dog. These things are hard for a machine to do, exponentially so. There's a reason why fully autonomous cars are only tested in CA or on closed tracks at this point: they're fucked if it e.g. rains or snows, if there's dirt on the pavement, or even if someone repainted the road markings (the cars use the pre-recorded view of the pavement to determine the position of the vehicle within the lane).
So e.g. Google does it so that it can pretend it's about "moonshots" rather than just selling your privacy to the highest bidder -- that helps with hiring. Hotz is doing this because there's free money and fame to be had, and to be fair, it's a pretty cool garage project. Uber is doing it to pretend it's "just as good as Google", because luring engineers solely with money tends to attract the wrong kind of people.
Rather than "self driving" cars, I'd love to see a human-driven car with advanced safety features that's darn near impossible to crash or run someone over with. I'd be willing to pay $10K just for that, over and above the price of the actual car.
The problem with the NHTSA... Is that they both write and enforce the laws.
Congratulations on failing to grok our legislative and regulatory system! Well done, AC! ;)
Anons need not reply. Questions end with a question mark.
So-called 'self-driving vehicles' are a subject that must be approached with the utmost of caution considering that a defective system of this type could cause massive loss of human lives. Companies developing such systems must be required to at least as cautious, and preferably doubly so, than medical device or pharmaceutical manufacturers are in the testing of their products for safety.
Would you prefer he said "write and enforce the regulations"? It would mean about the same. Or are you being picky, in that the actual enforcement would be done by police?
The regulations seem very reasonable at this time and from this distance, but the PROCESS is not reasonable, and arguably unconstitutional. That something *like* this needs to be done doesn't mean that doing things in this way doesn't pose severe problems in many cases, and it doesn't mean that the process is constitutional. Mind you, I would believe that the supreme court would (or has) decided that it does, but they make obviously invalid arguments to justify actions that appear to be needed. Much of the time I agree with their intent, but I often find their arguments to require extreme contortion of the language in a way that I find totally invalid. But I'm a programmer, not a lawyer. I didn't study how to make yes mean no and conversely.
I think we've pushed this "anyone can grow up to be president" thing too far.
It is true that 90% of the problem of autonomous driving is solved. It's also true that the remaining 10% are exponentially hard and won't be solved in the next 20 years.
Nice to see someone commenting on this subject that isn't so blindly enthusiastic about it that they adopt the potentially fatal 'what could possibly go wrong' attitude towards it.
Unlike so many technology products, self-driving cars have the potential to cause massive loss of human lives if they're defective or fail. Even 99% isn't close enough, it must be 100% perfect 100% of the time, or it's not safe enough.
Of course someone will now chime in with something along the lines of 'humans are not competent to drive and any machine would do a better job', but I disagree, if the human in question is one of the people whose life is at stake if the system fails, then that human being should have the right and the ability to be the final backup to the automated system. Since asking any machine to be 100% perfect 100% of the time is not attainable in the real world, then it must be made possible for a human driver to take over from the automated system at any time.
I actually disagree with 99% assertion. The relative importance of problems matters here. To succeed, this technology needs to be _substantially_ safer than humans _in all cases that matter_, not in all possible cases. That having been said, even with a relaxed problem statement like this, with today's technology this is a nearly insurmountable goal.
Most people have no idea what "information" is required when approval is required for a life critical system involving software.
You are absolutely correct. Having worked for a diagnostic medical device manufacturer for some years, I am cognizant of what the dollar cost and time required is to perform FDA qualification testing for such devices before it may be certified for sale in the United States. It's extensive and comprehensive, and we weren't even building a machine that could potentially endanger someone's life if it was defective or failed in some way. A 'self-driving' vehicle, and especially a 'driverless' vehicle has the potential to cause massive loss of human life if it is defective or fails. It's fitness for use is totally binary; it either works perfectly, or it fails completely, nothing between the two is acceptable. Meanwhile all companies attempting to develop such systems are investing many many millions of dollars. I can just imagine how much pressure they're under from their boards of directors, stockholders, and investors, to bring these products to market and start generating profits from them. Unfortunately for all of them however profits cannot come before safety in this case. If your smartphone or computer crashes that's disappointing; if a dozen people die in an accident because a self-driving car system has a flaw or an unforseen failure, then that's completely unacceptable and someone should be going to jail over it and/or paying huge amounts in settlements.
My point is that if it's not going to be 100% reliable 100% of the time then you can't have a self-driving car that has no controls for a human driver. Knowing how systems are developed and the challenges in creating such a system I can't see such a vehicle ever being allowed for public use and would certainly never set foot inside one unless I was feeling suicidal. That handful of percentage points that it can't handle means you must still be competent to manually operate a vehicle and pay attention at all times, and the vehicle must have a full set of controls for a human operator that are availble and active at all times. Of course one of my fears about technology such as this is that while humans will still be required to be competent drivers, these systems will make them lazy and their driving skills will atrophy, and when the time comes that they are needed to control the vehicle, they won't be able to do it anymore.
A failwhale of tweets.
Socialism: a lie told by totalitarians and believed by fools.
Google actually did research on this. They've found that, counterintuitively, it's dangerous to have a "mostly" autonomous car that can yield to a human driver at any moment. People tend to underestimate how much context they're holding in their head as they are driving. Acquiring that context at a moment's notice is just not really possible, so people tend to make stupid decisions if they weren't already paying attention. So it has to be either 100% hands free, or it'll have to be sold as a glorified driver safety system. That's why the cute little self driving cars that Google uses on its campus don't have a steering wheel at all.
Fleet has the advantages of not being associated with the site (tweeters are called twits by many), and of being a popular home enema.
This works because the outcome of either fleet is a lot of crap...
While we are at it, if someone is smitten with a twit, are they twitten? Or if you want to go with tweeter, are followers twetties? Warner Bros. may have an issue with that one...
You have the right to remain sentient. If you give up the right to remain sentient, you will be elected to public office
But I maintain that it's suicidal to get into a 'self driving car' that has no way for you to control it, and I know for a fact that I'm far, far from being alone in that. You won't be able to sell the public on a so-called 'self-driving car' that has no way for them to control it with their own two hands and own two feet, and they won't be satisfied with some sort of voice control, or even a big red 'Emergency Stop' button. It'll have to be a 'glorified driver safety system', then, and I think the vast majority of people will be perfectly fine with that. Besides which there will always be circumstances where an automated system just can't cope, and for those situations there has to be manual controls so a human being can maneuver the vehicle, and video game controls or some sort of keypad just isn't going to cut it, either. It needs to be the same traditional controls we've had all along. So it may as well be a sophisticated 'cruise control' feature that also happens to help keep you from getting in collisions or veering off the road. Also realize that where in 5 to 10 years you might be able to buy a vehicle with some system like this, it's only going to be on high-end luxury cars as an expensive option, not on your average everyday economy car that the vast majority of people can afford. There's no government mandate or law you can trot out, outlawing manually driven vehicles, that would ever make it into law; you're not going to tell 200,000,000 Americans that they have to throw their cars away and buy an expensive new one whether they like it or not. At best you're looking at 20 to 30 years before this makes it into the mainstream, and that's being very liberal in my estimate. In the meantime people will still drive their own cars. I'd recommend we institute reforms in driver education and training to improve driver competency in the meantime, things I've heard about how they train new drivers now makes me wonder what they're thinking.