The problem you then face is that any of those could be forged / faked without some kind of way of securely validating the message in some way. You could cause absolute chaos by driving down the road broadcasting false messages. It's a little harder to hack and modify traffic light signals, for example. But we've also seen hackers screw up Tesla cars by sticking stuff on the back of their car to deliberate mislead it based on vision.
Even without self-driving cars, an "attacker" can go into a theater and yell "fire" and cause a stampede.
They can get a high-viz vest and clipboard, and stand in intersections directing cars to take detours they don't need and holding up traffic.
My point here is that society has a lot of trust baked in. We trust people don't just yell "fire" without reason. Just because it's FSD cars doesn't mean people will start broadcasting the equivalent of "fire" constantly. It's already easy to cause accidents.
The consequences of potential attacks on centrally-orchestrated traffic are a lot more severe. Hack the control node, and you can stop traffic nation-wide. Or cause mass accidents that overwhelm first responders. And they can be executed by anyone, anywhere in the world, for a cost within range of many medium-size corporations (let alone nation states).
I won't comment on the challenges of the approach Tesla et al. are currently taking, but I don't think central control is the panacea commenters in this thread are making it out to be (and I'm personally glad this isn't the route we're pursuing).
It’s like arguing that we can’t possibly build autonomous cars because then someone might turn it into an autonomous bomb.
Keep in mind that solving this is “worth” about 40,000 lives a year in the US - nearly $1 trillion in economic damages a year in life and property.
Bad things can always be done with good tools. As always, you provide layers of protection that make sense and in the end must rely on the underlying fabric of civilization to persevere.
I’ve personally tested creating a fake toad sign and my tesla reads it as a real sign just fine.
I wonder what happens to autopilot on a 70mph freeway when it encounters a 5mph limit sign…
It immediately reduced speed from 60mph to 25mph .. aggressively.
This falls into a pattern of Tesla autopilot/NoA where it just doesn't seem to have much memory or foresight.
For example the car is driving itself on the highway, it knows it's been on the highway, for 20 minutes. I am not even in the exit lane, it knows what lane I am in. How could it think I am suddenly on the local road below the highway based solely on the GPS pin movement in the span of a second, without having moved to the exit lane and gone down the exit ramp?
For an example of lack of foresight - the car will happily speed towards an obvious semi-distant slowdown right until it needs to aggressively break from 60mph down to 30mph as it approaches following distance of the nearest car. I also find it can get really weird in stop&go traffic, not easing into speed, down to a stop very well as if it has only GO or STOP.
Oh wow. This happens often with a car-mounted GPS (or on a phone) and it's pretty annoying. Sometimes the GPS instructs you to do a U-turn at the next available fork in the road, and it takes a moment to understand what's going on.
But in a self-driving car it's terrifying! And absurd.
Any reason routine crypto methods would not solve this? Seems like one of the easier parts to me.
The real concern would be whether someone can engineer a terrorist level mass scale attack but as long as it requires physical tampering that adds up to a tremendous amount of work. So if the signalling is largely burned into fixed infastructure it eliminates a lot of that or at least sets the bar high enough that its probably more work than various other types of attack that are likely to be just as impactful.
Sounds like a job for...blockchain.
Yep, I talk to people working in traffic engineering, and their mindset is always building new road tech and road-side and cloud infra to support autonomous driving. They have no expectation of fully autonomous vehicle without road and infrastructure assistance.
And from historical perspective, the coming of automotive and the replacement of horse and other animal carts, are exactly facilitated by the road transformation; which has been the single largest scale infrastructure in human history.
It makes no sense that an even bigger transformation of the vehicle would require less drastic road transformation.
The very marginal benefits of laying road vs track more or less disappear when automation in play.
Is is really worth maintaining the ability to go off road/track when you’re not even driving anymore?
This idea is not new and it mostly applies to freight convoys but I think it also has merit for ad hoc passenger car convoys on long highway trips.
I think the idea is to get rid of the roads and have the robo vehicles travel instead on tracks. That increases fuel efficiency and bypasses a lot of AI challenges.
It’s the low-key case that Elon will be remembered poorly (like Robert Moses’ rapidly degrading legacy) for
1) Having the wrong vision for EVs (but successfully executing non it anyway)
2) Making space travel cheap (thereby increasing the amount of carbon energy dedicated to it) without really improving an average human’s quality of life
Fortunately thats not a lot of people.
When electricity was being rolled out, Westinghouse and Edison didn’t bellyache about not being able to provide electricity to rural areas. They electrified all the cities.
Rural areas will just never be able to pay for modern infrastructure. And… thats ok, its not a lot of people.
This is a thing in Europe, and even some US cities - my Audi has traffic sign recognition and when at a compatible intersection knows what the light is at (by radio, not by light), and how long until it changes (will show a countdown in seconds til the next light change).
Are our roads really? Most in cities over a certain age are just haphazard relics of times gone by, and don't get me started on "stroads" which are good for nobody
This aspect of FSD has always fascinated me and I'm a little surprised it doesn't get more discussion. Meeting halfway. At what point could/would/should FSD influence the environment around it?
For example - a poorly painted road sign*. Tesla/Waymo could say "We cannot support L5 FSD on this road until you fix this sign." If it meant a step forward in autonomy, Tesla/Waymo could even offer to share the cost of that improvement!
There are a million reasons why implementation of that would be problematic. Costs and incentives would be all over the place. But I am more interested in the framing: The machines are the ones that need to adapt. Which is essentially hoping for continued hardware improvements or a spaghetti mess of if/else statements. ie "do this weird thing if you see this other weird thing in front of you". Can we get rid of the weird thing and avoid the engineering challenge altogether?
* Yes, this is an overly simple example. Some environment changes could be so large that they would require a full redesign of a city/buildings/traffic patterns. But surely there are classes of improvements where some are easier than others.
what's missing is combining this kind of "human concept relations" model (language, rules, minimal reasoning, text encoded human preferences) with perception, and safety (which means that the model should know that if other cars are driving just fine in front then it's unlikely that the road is on fire, or that the low certainty crack in the road is okay if two other cars already went over it unimpeded, if the road marks and the signs are inconsistent, but other vehicles have formed a slow but consistent pattern of traffic then that's the local ruleset, and so on)
it's still a very hard problem. and the required amount of compute is still bonkers, the required amount of data and training is still absolutely huge, and the whole problem of safely disengaging, handling the asleep/drunk passengers (likely target audience after all)... are all hard problems too :)
Or even cycle? I hear great things.
A part of the reason people wish they had FSD so badly is because they want to be rescued from this fundamental failure of NA-style urban planning that necessities driving, all the time, across both short and long distances.
It doesn't work like that in most countries; small cities and towns are navigable on foot and public transport.
If you could replace double-decker buses that arrive every 15–30 minutes with self-driving minibuses that arrive every 3–5 minutes, that would be great! (for everyone except the bus drivers who lose their jobs)
Actually, I suspect this makes self-driving a particularly _bad_ solution for city buses; getting into the bus stops takes some manoeuvring, particularly when there are other buses there.
One place that self-driving buses could be interesting (and indeed there are already a couple of systems like this) is on fully/near-fully segregated lines, where they don't have to deal with human-operated traffic. Another would be small towns, but you're looking at full magic level 5 at that point.
I’m envisaging a system of minibuses, either AI-driven or at least dynamically directed by a central control system. People would use a phone app to book journeys; the app would tell them where to get on and where to transfer, and the central control system would optimise the fleet to get everyone where they need to go.
With much smaller buses, and electronic tap-in rather than cash payments, stops should be fast enough that buses can just queue up in order at each stop, hopefully mitigating the parking difficulties you mentioned (modulo breakdowns, medical emergencies, etc). Likewise, if transfers are fast and easy, hopefully they’d be less objectionable to travellers.
Requiring a phone isn’t ideal as it limits accessibility and privacy. There could also be pre-printed tickets, with QR codes that you scan at the bus stop to see the route info.
Why minibuses and not just taxis? I suspect there’s a good balance to be made between efficient road usage (buses) and efficient routing for each traveller (cars).
I’m also envisaging that if this system were to take off, personal cars could be gradually removed from city centres! Again, that makes life easier for the AI vehicles.
This is all pie-in-the-sky stuff, I know; but I do feel like there are ways we could radically improve city transport mostly using existing roads, rather than building new rails or tunnels.
I heard the other way - people leaving cities for suburbs
Things like traffic signals that actively communicate their status to nearby robot cars (more than just a red lamp that can be occluded by weather, other vehicles, or mud on the camera lens). Or lane markings that are more than just reflective paint, but can be sensed via RF. Rules around temporary construction that dictate the manner of signage and cone placement that the robot cars can understand. The cones might have little transponders in them, I don't know.
But without a massive leap forward in AI capability, our current road system—optimized for human drivers over the past century—is not going to work.
If we can't make the cars just as smart as an alert and capable driver, then maybe we need to meet halfway and make the roads a littler "dumber" (simpler) to accommodate the robots.