I really don't want connected cars. It introduces way too much remote attack surface where there was zero before.
We can do that today with remote start on a hot day. And on a cold day, it remotely activates the heater.
Isn't there already a significant attack vector ?
And the pressure is high for makers to bring more of these sooner than later, so having a more public and wider discussion on what this means on the security side is I think beneficial. Right now they're burying their head in the sand.
What if you could do 100mph with perfect congestion control in the e-lane?
Just imagine a world where rush hour didn't mean slowdowns. Oh and also get rid of the onramp metering; it's worse than useless when traffic can't zipper and accommodate the influx of new vehicles.
The rulemakers would allow your car to drive at a MUCH higher speed than allowed by "human" driving.
and it would maintain the speed.
The cars would perfectly merge in and out, maybe including perfect metering or buffering of cars exiting.
They would drive at a specific distance from each other and speed up and slow down synchronously.
oh, and don't charge for it.
If you want to own one for exclusive personal use, you'll be doing so in an environment that's increasingly unsympathetic to your needs as somebody with a fleet size of one and a customer base of one.
Tokyo is of comparable size to LA and they don't have these issues. LA used to have the largest streetcar network in the country but they stripped most of it out circa WWII. See also NYC, London, etc.
Even if most people would take the tube, that still leaves millions who need to drive into LA for work. Since everyone has a backyard and the population density in the suburbs doesn't come anywhere close to Tokyo's, it's neither economical nor practical to build public transportation options out there.
> Tokyo is of comparable size to LA and they don't have these issues. LA used to have the largest streetcar network in the country but they stripped most of it out circa WWII. See also NYC, London, etc.
Used to, but those days are gone and it's now 100x harder to reverse that decision. Los Angeles is currently spending the most money out of all the cities in the US on its metro system in preparation for the 2028 Olympics, and even that is still scratching the surface.
Teslas communicating with each other isn't using an open standard that can be used by other car companies, nor is Waymo's.
Germany has so much better driving norms on its highways since the lack of speed limits enables those who are fast to force those who are slow to stay in the right lane where they belong.
In an oversimplified system where Car A broadcasts "I'm braking" allowing Car B to slow down and avoid a collision, the attack vector is a simulated "I'm braking" message that causes car B to slow down/stop even though Car A is not braking (or may not even exist).
Broadcasting current mode of operation doesn't really seem connected in the same way to me. Sure, it might be a way to "attack" another car by sending the same signal, but that's totally different from someone accessing the car remotely for other purposes. If you fake a hard braking signal, to my car, then my car will respond by slowing down and then transmitting that as well to other cars.
It's not safe. It never will be safe. Ever. Self driving cars should have absolutely zero networking capability, at all.
Anyone saying otherwise is ignoring te reality of software development history, and extremely naive.
Thus if your car needs to communicate with anything for "system updates", it's been designed wrong. People had cars for decades, with digital control systems and zero networking capability. It's literally not required, at all, to develop, maintain, or have a car that operates perfectly.
One of the main problems is that cars literally have too much software onboard. There is no need for an app store for a car. No need for networking. No need for update-to-date info. None. You have a smartphone, and that can connect to an entirely isolated screen in the car, if you want maps displayed for your own edification. Anything networked, eg bluetooth, etc, should be entirely isolated from the rest of the car.
In terms of self-driving, updates can be applied manually. USB sticks aren't that uncommon. Dealers are available. The amount of times new city streets are created is extremely rare, and you can always close-destination and direct manually beyond that point. Cars can then remember a location, and draw in a street as if finds it, thus enabling easy return.
The truth is it doesn't matter how "convenient" something may be, you don't just brush security and safety aside to do it. You don't make people's lives easier, at the expense of safety, security, and so on. You just don't.
And that's what every networked car represents. Brushing aside safety for convenience.
I've been waiting for hackers to remote-hack the battery charging module in cars for quite some time. Depending on the unit's configuration, some car models could be hacked to all explode and burst into flames due to overcharging, at specific times.
How would society respond, in 1 in 10 houses caught fire at 2am on the same night?
* There'd be fires everywhere, and those fires would spread, as there are not enough fire departments to deal with even 1 in 100 houses catching fire in a night
* Massive amounts of infra would be compromised
* Massive amounts of transportation capability would be gone
Society would be devastated. It'd be worse than an air bombing campaign.
Yet I'm willing to bet there is a path from most car's networking -> charge controller, along with it being remote flashable with a new firmware too.
Madness. Stupidity. Insanity.
Other discussions are about how people coordinate cars in a group. This is ripe for trouble, even with just people messing about. People would game this system to push other cars out of the way, using it to gain pseudo priority. Teenagers and malign actors would cause all cars in a swarm to emergency brake, but sending emergency brake intentions. Cars would be manipulated into running into each other, or into guardrails, after being sent "emergency swerve" info from cars in front, or "I'm beside you but I'm emergency swerving into your lane!" messages.
If anyone looks at the current state of almost all lane-keep tech, it's a joke and dangerous. Forums are repleted with "turn off this functionality" panic messages, as people are having to constantly fight their cars to keep them from doing very dumb things.
Real self-driving is at least 20 to 30 years away. Even those at the forefront (like Waymo) are not self-driving. Instead, they've super-mapped the areas that Waymo operates in, warm climates with low amounts of rain, no snow, city areas with unchanging landscapes and an immense amount of cues and markers.
Take a Waymo to an unmapped location and it's useless. That's not self-driving.
Take a Waymo to a Northern US rural area in the winter, where there are no road lines to be seen due to slush and snow on the road. It will fail completely and utterly. That's not self driving. Any local driver handles that daily without issue.
Waymo still has a phalanx of people monitoring all aspects of how its cars are performing, ready to step in at a moment's notice. They get edge-case stuck in dumb places, do weird things (like waking up entire blocks of people due to honking senselessly all night long), they are not self-driving capable.
Instead, they are "we mapped all these streets a billion times over, and with human intervention and the absolutely perfect weather in SF and the US South, can get a car to kinds sorta almost self-drive", but "It'll never work in the rest of the world".
By the time true self-driving is a real thing, it'll be hand in hand with AI that is capable of handling "I don't have that address" and "Show me the closest address" and "where do I go from here", where you say "Go down two streets and then turn left" works.
We do not need networking in our cars. The internet is a poison.
One side note.
This is an excellent business opportunity for someone such as SpaceX. They could create entirely isolated vlan, non-internet traffic for car manufacturers far more isolated than anyone. Such traffic could be isolated and never have direct connection to the internet, which would reduce the attack surface dramatically.
It would still be dangerous. Just less so than the absolutely absurd and ridiculous scenario where we have cars connect to the most hostile environment known, the internet.
To end this blathering and spewing, I'll say this. Would you want brain surgery done by a robot that is receiving instructions over the internet? Then why do you want tonnes of vehicles flying down the road at 70mph being connected to the internet?
First, the idea that you rarely need to update maps is wrong. The minimum you can really get away with is monthly. Weekly is preferable. Daily is the standard, with serious pressure from operations and government relations support up-to-the-minute updates to handle things like avoiding emergency services routes, route obstructions like downed trees, traffic updates, and unannounced construction zones detected by another vehicle.
Secondly, you missed the "data offload" part of what I said. One of the main limitations on operating time for autonomous vehicles isn't charging, it's running out of space to store collected data. This includes data that will be used to update maps, detailed logs of how the vehicle is running and what errors sensors are encountering, as well as the basic sensor outputs and analysis results (e.g. where other vehicles are in space). This is often terabytes after compression and log reduction. Good luck loading that to a USB drive in any reasonable time.
Thirdly, good luck making USB devices secure against evil maid attacks. It's a hell of a lot easier and there's a hell of a lot more bandwidth available with mutually authenticated wireless APs.
As for taking Waymo to a northern state, they've been testing in Tahoe for years and they used to test in New York. This coming winter they'll be doing heavy testing in both of those as well as Northern Michigan (including UP).
Waymo are autonomous. They're not on rails or anything else you've heard. The computers are given a destination and route themselves to it, handling all driving tasks along the way. When remote assistants step in, all they can do is augment the robot with new observations. It still has to do all of the driving in light of that new information.
No, it isn't. Many people do not use Google Maps, or any maps like technology as they drive about. They rely upon road signs, and signs saying "Maintenance from Dec to March", and so on, along with 'detour' signs. Cars can easily read road signs these days, and that can be expanded. Further, people can put in their own "bypass" route, which could even be augmented "For the next week". There is no requirement to update often, except for quite literally made up requirements, beyond the requirements humans have.
A true requirement for full self-driving, is the ability to indicate "STOP!" or "Let me off here", or for example "Take this street to bypass this mess in the future", whether verbal of via a console. Self-driving doesn't mean you don't tell the chauffeur your preferences, outside of it reading detour signs. In fact, it's a requirement to take directions from a human in the vehicle, as to destination, route, emergency stops, and so on.
And this highlights my point. Dozens of ways to have maps updated less often, very low impact, but suddenly "We need to connect to the most dangerous thing for any computing device, the Internet".
--
In terms of data offload, it's not required. If it was an actual self-driving car, it could drive in areas with no mobile or cell service, and yes many such places exist. It could be owned as a self-driving car, where people live "off the grid" entirely. What would happen then? Would the vehicle cease to function due to a full disk? Crash? That's very, very poor design.
And to that end, all of that data is essentially not required for operations, but for debug and improvement. And this highlights how beta-ish this tech is. You don't need daily logs and endless updates on something stable. The core components of auto-driving vehicles should be in perm-maintenance only mode, with on additions or changes, that's where stable code ends up. No changes.
--
In terms of evil maid attacks, seriously? The mega edge case, compared to the lunacy of placing a device online? Online is dumb, dumb, dumb. It's unsecure beyond comparison. If you have anything you care about online, you're doing it wrong.
--
In terms of testing, yes I'm sure they are testing. Yet they're not testing anywhere near their operations in SF, and the number of Waymos on test drives around Palo Alto dwarfs anything in a snow laden environment.
Testing doesn't mean they've managed to make any headway against blizzard conditions, against snow on road, against unmapped roads with snow on them, against the car icing up, on and on and on. Testing != working.
--
And in terms of your comment about remotes, I am unsure of the validity of your statement, but regardless, helping it determine landmarks (aka new observations) is indeed my point. SF is the most mapped area on the planet. And yet people still have to step in and provide guidance.
Self-driving is a joke at this point, the best is barely 4 as per J3016, and this chart is a best-describe:
https://images.synopsys.com/is/image/synopsys/levels-of-driv...
Note the 'geofencing is required' in this specific definition, which is key to how immensely limited Waymo is, and I'd take this further and say Waymo is a '3.5' at best, with 'human is required' being offloaded for emergency intervention/stops if required. (Don't try to tell me Waymo doesn't do emergency remote stops.. it happens. And 'human is required' doesn't only mean 'the one in the car').
Couple this with the fact that it's extremely easy to make something perform correctly some of the time, difficult most of the time, and an immense feat of incredible engineering for all of the time, even when all means a whole slew of 9s, not 100%.
True self-driving without all sorts of conditionals is a minimum of 10 years out, most likely closer to 20 years, and I wouldn't be surprised if 30.
People have such rose coloured glasses on this front.
We don't have cell towers everywhere so there are 'blind' spots for miles.