To solve the self-driving problem we need "smart" A.I., which means we have to approach it with systematic engineering, and the solution will probably involve some combination of better sensors, introspectable neural nets, symbolic A.I., and logical A.I.
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.
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.
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…
Any reason routine crypto methods would not solve this? Seems like one of the easier parts to me.
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 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.
Edit: I have a whole mental model for other drivers and different approaches for them. Someone driving like a grandma? Pass when available. Nervous/erratic/lost driver? Keep extra distance then pass as soon as possible. Aggressive driver? Relax, give some space and let them get ahead. And so on. I get that stereotyping is bad but ignoring the subtle signals other drivers give off seems like it would be myopic. An AI that doesn't anticipate what others will do on the road will always be reactive rather than proactive.
I maintain the position that if self-driving cars handle the most common situations as well as average human beings, there won't be a strong drive to make cars that drive significantly better than humans. Cruise and Waymo are collecting a limited version of that dataset right now. It's unclear how many deaths because some details are being kept under wraps.
