Toyota Lifts the Veil on Its Guardian Driver-Assist System
spectrum.ieee.org
spectrum.ieee.org
He pointed me to GeorgeHotz YouTube channel who developed in the open for CommaAI (now shut down), this company got 8m in funding (small cheese to a VC, but still showed there was interest in the project) and he was just googling answers and training his model on one single video he downloaded of the internet.
The 'play loose and fast' with the rules approach won't fly with self-driving and the car companies are looking to build something that works in the field with as close to imperceptible failures as can be realistically allowed.
The incumbents are building it right, but slow. And I would much prefer that approach if it leads to increased safety
Until it's done correctly by any side, nobody is "right" and the way it seems or not is a PR thing. At the very least, one defending the traditional car manufacturers' approaches would at least concede the impetus might never had arisen, or not for a long time, with their closeted and often-unrealized research methods. Especially if it's counter to the traditional income funnel.
The reality is that Silicon Valley is very very good at moving fast and building very high tech solutions very quickly. The car industry isn't exactly known for being fast movers and they were very slow to get on this bandwagon.
I'm sure Silicon Valley has a few things to learn from the incumbent car industry too but let's face it - the car industry is mostly just desperate not to have their lunch eaten at this point.
If I were in your shoes, I totally would have gone there.
Wait, are you serious? But I mean, I thought the whole point of driving a car in the first place was that you could "move fast and break things!" So, clearly Silicon Valley has the most appealing approach!
From every source I've read that's publicly available, SV IS eating the major automakers' lunch. An anecdotal claim with zero evidence isn't enough to convince me otherwise, nor should it convince you.
What lunch are they eating?
In short: the likely explanation for those failures, regardless of bad coding practices, were most likely 100% human error
Are you saying that Comcast, a company that has a monopoly on internet access, has zero incentive to work on increasing bandwidth speeds and lowering prices for end users?
You see how wrong and colossally stupid your argument is, saying that corporations with essentially a monopoly, without incentive, will just happily increase the quality of their product?
Your argument is so wrong and colossally stupid that I feel like Humanity as a whole, has taken a step back due to the ignorance that has happened here today.
I really like this idea but I feel this is the key in getting it right. Several video games achieve this feeling of seamlessness when translating raw input from a controller to in-game actions. However this does occasionally go wrong and can be a frustrating experience in video games (especially if you're playing online competitively) but in the real world it can have much more drastic consequences.
Also, given the complexity of the real world, it's somewhat scary to think of a scenario where you're actually trying to take a drastic action on the road but are prevented by the system from doing-so because it feels that the action is unsafe.
Its nice to dream about ideal future, but if lives rely on reliability, it either works 100%, or shouldn't.
People like to go fast in their cars without having to be as skilled? It’s not exactly a mystery.
Do they want to be skilled or do they just want to have some easy fun?
If I could show up at the Nürburgring and hot lap a 911 on nanny driver mode I'd do it in a heart beat. Until then I'll just stick with not getting into Berghain.
Even a manual override would be insufficient because by the time you realize it won't behave correctly, it's too late to apply the override and then take corrective action. I would probably not trust this system, in my hubris.
https://www.automobilemag.com/news/2018-volvo-xc60-arrives-s...
Is automation paradox a real issue? Absolutely. But at present, all the evidence seems to indicate that the people automated cars will save far outnumber the people they will kill.
Focusing on the dangers is alarmist and misleading. The 737 crash is a good example really - I've never heard anyone with experience in the field suggest anything other than that the automation makes planes safer. Focusing instead on the fact that the automation isn't perfectly safe only makes sense if you're a media company looking to scare people.
Heck, my car has adaptive cruise control, and I already find myself wandering sometimes, assuming that it will slam on the brakes if someone gets in front of me. And I'm well aware of the automation paradox and its dangers.
The current evidence suggests that semi-autonomous cars are on balance safer than purely human driven cars. The evidence isn't conclusive, but I've literally seen no evidence presented to support the case that semi-autonomous cars are more dangerous - at best I've seen people argue that we can't yet trust the safety claims. Teslas have been on the road long enough now and in large enough numbers that if they were really more dangerous than non-semi-autonomous cars, I think we'd have seen some non-anecdotal evidence by now.
The reasonable presentation of the issue would be "semi-autonomous vehicles likely make people safer, but there are still dangers and people should pay attention."
Instead, the story you're telling is "Tesla drivers are dying, semi-autonomous cars are dangerous, and you should be scared."
You must see how this is misleading at best, and likely downright counterproductive from an overall safety point of view.
Tesla's so-called "autopilot" looks shockingly dangerous and reckless in comparison, and may hinder acceptance of autonomous driving in the long-term.
It's immediately obvious to someone who can't drive a manual that they need to switch to a vehicle with an automatic transmission.
Someone who learned on or became accustomed to a vehicle that transparently corrects all of their mistakes could easily end up in way over their head before realizing that the car they're driving doesn't make up for their lack of basic lane keeping skills / etc.
I have had one incident where I had to swerve onto the wrong side of the road to dodge a car that suddenly pulled into my lane from a line of parked cars (with no indicator and no warning). I was lucky there was no oncoming traffic because there was no way I could have stopped without hitting the car that pulled out.
I didn't stall then but I also didn't have to come to a complete stop, like if a kid had jumped onto the road and I had nowhere to go and had to slam the brakes completely on.
I don't think stalling is really a major safety issue though. About the only example I can think of where you'd really have to accelerate to be safe is when your car is stalled on railway tracks, and in that case you'd be unlikely to have had to do some maneuver that landed you there.
99.99% of the time slowing down is safer than accelerating.
In all of those cases you're dangerous for maybe a day while you familiarize yourself with the new vehicle and most people behave very, very conservatively when driving something different than they're used to. Nobody is passing their test in a Miata then renting a box truck and saying "yes, I can totally take this hairpin just as fast as I would in a Miata". The biggest danger from people switching from small to large vehicles seems to be bumping things in tight spaces (parking lots) and with manual the danger is rolling back too much on a hill start or stalling it when trying to pull into traffic. While certainly expensive and inconvenient I don't think those sorts of accidents are cause enough danger to the public or are common enough to prohibit people from driving a manual or renting a box truck without a special license.
I think people being unfamiliar with what they're driving is an edge case a lot like mechanical failure, everyone hand wrings over it, some jurisdictions take steps to try and prevent it but in the real world it's just such a tiny fraction of accidents (mostly minor single car accidents at that) compared to people being stupid that trying to prevent these edge cases is a waste of time and money. Resources are finite and they would be better spent elsewhere.
And no, driving a bigger(especially larger than 3.5 tonnes) vehicle is not just about the vehicle being physically larger and harder to park. That's why the European categories for driving such large vehicles(C and D categories) have very different theoretical tests than a regular B category for driving small(<3.5 tonne) cars. There's a whole host of laws saying where and when you can drive such a vehicle, and speed limits are different for them, regardless of what the speed limit signs say. Part of the test for C category includes safe loading of a heavy vehicle - like in your example, if you rent a box truck and put a 2 tonne pallet on one side, you're going to have a bad time. That's why the training is important.
I would actually be very pleased if in the future there is a simpler test for driving a "clever" car, but which doesn't convey a regular license.
If you're unable to meet the current standard, you probably shouldn't be allowed to walk down the street unescorted.. let alone drive a car, regardless of the "cleverness" of that vehicle.
Jet fighters (also passenger ones) nowadays are flown by computers, you push the joystick to tell it where to go, and the computers control the flight surfaces to follow your request but not beyond the point that it would fall out of the sky. And TFA seems to suggest the system will prevent crashes, so it would probably stop you if you e.g. tried to swerve into an oncoming truck.
Regular planes are inherently stable so they can fly closer to a glider.
I just want a car.
I got one that wants to take over steering, disturbing me as I drive. I disabled that. It also wants to slam on the brakes whenever it gets confused. That misfeature must be disabled every time I start the car. Then there is the built-in cellular data and GPS and the microphone and at least 3 cameras. Yes, they are tracking me, and if they don't do worse already then it's just an involuntary software update away.
Some day this will be hacked, and millions will die. That will be a memorable day.
I wouldn't know where to find the microphone.
As for voiding any warranty, should a problem arrise, it's up to the manufacturer to prove that whatever you did caused damage. Just taking apart and reassembling a console correctly will not void your warranty. "Warranty Voided if removed/opened" labels are not legally-enforceable in the US.
Just look for Youtube videos on how to take it apart (or look for repair videos for other items in the console).
Then again, I do share your concern about hackability. Just a single grand catastrophe would turn the scales over for a long time. And nothing suggests that car software is of high quality currently.
>Do you drive while text messaging
Amazingly many people do, even after being told that it's about as dangerous as driving drunk.
Can we all trust the car companies to do their due diligence and have all their software audited. Do we trust their proprietary algorithms?
At least when i am driving myself i know what is going on in my head. Yes it might sound a little conceded, but who do you really trust.
Also as a side mark (to get me downvoted :) ), i can understand automatic braking and some of the other things, but if you need the machine just to help maintain your lane for you...you probably should not be driving (i seriously work with someone who praises their car for helping them not drift out of their lane).
This is not actually an argument against putting lane keeping assist functionality into cars. Yes, distracted driving is bad. But there's no way to completely prevent it, so the auto industry should do everything they can to mitigate the harm.
If you hold peoples hands on stuff they start slacking, they become dependent and cannot do something without assistance. My kid plays a FPS online on his xbox one, he started off doing ok and getting better. He found out you can turn on aim assist and got really good. If he turns off the aim assistance, he is worse than he ever was.
I am not disagreeing with you that lane assist, now that is exists, might not be useful. I just feel we are moving to a world like the people on the ship in the movie Wall-e.
I trust it as much as you trust a car to not have dangerous mechanical problems.
> Do you drive while text messaging, commenting on Hacker News, or watching video?
No. But distractions happen. Pretending otherwise is hubris.
> Are you perfectly fine with your whereabouts and driving style being known and sold, for example to insurance companies or toll collectors or the police?
These are addons that don't come with the car. With this, I've lowered my insurance costs, and now going through tolls is much easier and less dangerous.
> Are you using the features to compensate for any kind of disability?
Being human and knowing that I'm not perfect.
> Would you be fine with a car like HAL 9000, saying "I'm sorry muzika, I'm afraid I can't do that" to you?
That doesn't happen.
> This is alien to me.
That is clear. Judging by your questions, you have absolutely no idea what is in these cars. It's good thought that you are aware of your ignorance, admit to it, and seek to learn more. Hopefully the information I provide can help you.
I trust it as much as you trust a car to not have dangerous mechanical problems.
The problem with software is that you can't look at it to see if it's worn, you can't hear if it's about to fail, and it's not regularly inspected by experts to see how it is behaving. Also as programmers we are aware of hidden bugs that can only show up under certain circumstances. It's only a matter of time until an overflow or off-by-one error kills people.
Car makers and engineers have a very good idea how dangerous certain parts of cars are, and they have been built over many years to mitigate those weaknesses. So you wouldn't expect, for example, the five nuts holding a wheel on to all fail at once.
We programmers know that there can be far less certainty with software.
Even worse, others here have posted about the terrible software practices that were exposed at some manufacturers, and the reluctance by Jeep to react until the ability to crash their cars remotely was exposed at Blackhat.
I think it all comes down to how much trust you have and where that trust lies.
In my hubris, I'm going to have as analog a car as I can, and hold out without as many these features as I can until all the beta software bugs are ironed out. For mechanical failures, it took us many painful years to get to where we are now.
I guess what I'm saying is that I trust myself to pay attention and stay in the lane using the steering wheel more than I trust our current technology not to get confused and kill me by steering wrong. The reward just seems much less than the risk to me for that (and other) features. I don't trust automatic emergency braking to not slam the brakes on at a time when it will kill me either.
What scares me most is the perverse incentives the car manufacturers have to cover up their shortcomings. They have shown exactly how little they can be trusted to be honest with things such as VW cheating on emissions, exploding tires, faulty airbags and many many coverups to avoid recalls, etc, etc.
Also the remote access that is more and more prevalent in cars today. Hackers gaining access to the drivetrain functions of cars remotely is super scary to me. And when that happens, it will happen for every one of those cars on the road.
A DDOS of the morning rush would be very interesting at best. At worst, as bad as any terrorist attack.
Would you trust your life to the stability and security of even a major company's servers? I wouldn't either. There needs to be a LOT of regulation around these things, I believe, and we as a society should be able to decide exactly what we will accept and what we won't.
At the very least there should be rules about exactly how the systems in cars can be networked, and the security around those connections. The jeep hack was through a remote firmware update to the radio unit I believe. There are no important enough reasons to have a remotely-accessible radio directly networked to the drivetrain computer in my opinion.
It can't just be move fast and break things when lives are at stake. Everything I've heard sounds like it's the wild west at the moment.
The same goes for the milder forms of lane keeping assist; that only gets problematic if the system does enough steering that the driver can get away with taking their hands off the wheel entirely when traveling on a typical interstate highway. The assist systems that give the wheel a minor nudge and audibly alert the driver are an obvious net safety gain, especially for the innocent bystanders in the next lane over.
Many of the more advanced safety features that are being rolled out to the market these days can be done wrong and become a danger of their own. But when implemented properly, complaints against them are no more valid than complaints about anti-lock brakes and electronic stability control overriding the driver's intentions.
Example for lane following: I veer left a bit to avoid a bit of scrap wood with nails sticking out. The car counters this, steering me back toward the debris.
Essentially, anything cheap likely won't have this technology in.
It should be all or nothing control, at least that way people will not become marginally complacent, just out of practice (although most drivers can not drive for a month and drive again no problem).
Honestly I kind of feel like Toyota's still stuck in the 90s. They were ahead of the game but all they seem to have done since then is endlessly fine-tune and finesse the same vehicles.
If only the other mainstream manufacturers would follow suit?
https://www.usatoday.com/story/money/cars/2016/03/17/automat...
So actually less than four years, I thought it was for the 2022 models which would be released in late 2021.
edit: Seeing your later comment, you seem to think this is emergency braking. The video clearly shows this goes beyond emergency braking.
You can buy a 2017 Camry with emergency braking. It uses a laser and camera at the front and looks like a Camry. Compare the module to the Guardian sensor bank you see on the roof there. It's not for show. We will not see this sensor bank or these capabilities on every new car in 2022 and the article you posted doesn't remotely suggest that.
I think you're underestimating what Toyota is doing here.
There are quite a few vehicles that have at least some of the sensor capability required for this kind of functionality, but the hard part is the software/AI, not the cameras and LIDAR.
Mercedes cars do that. They also preemptively play grey noise over the speakers if the collision is imminent, to protect your eardrums - that's something I don't think any other manufacturer is doing.