San Francisco isn't a purpose-built lab, it's a real city that grew organically over centuries, like every other city. So if it works only in San Francisco now, it means it's 99% to working everywhere else on the planet.
> The world unfortunately is lot more diverse than that.
It is, but only at the surface level.
99% of the actual problem of driving in urban areas is the same everywhere on the planet, because all human cities are pretty much the same, because all humans are pretty much the same - once you start comparing with the rest of the universe, and not just with other people. San Francisco, Moscow, Cape Town and Prague are almost entirely unlike each other, yet they're all obviously the same when you compare them against ant hills or forests or coral reefs.
Self-driving is a physical reality problem, not social diversity perception problem.
Have you driven in these cities? I think it's quite a different problem!
But, there are a lot of places that are quite similar to San Francisco! Can we scale it to a thousand cities it'll still be revolutionary. And we then haven't even started talking automating transports! It's just taken longer than what people thought. The usual Gartner Hype Cycle.
> Driving in New Delhi is an entirely different universe than driving in San Francisco
Only if you assume driving is solved as a baseline. For self-serving, "driving" is the 99% of the problem; "driving in $specificCity" is just the cherry on the cake, a small part of the problem that doesn't even need to be solved.
Keep in mind that humans are the most flexible parts of any system they're in. When human and machine negotiate, human always yields, simply because the machine cannot. With increased popularity of self-driving cars, infrastructure and social norms will adjust to eliminate technical problems and cultural idiosynchrasies that the tech can't handle. This is how it always was.
That's sped up at least 5x. If you slow it down to real time, it looks no harder than Waymo videos in San Francisco. Remember that most AVs have full circle sensor coverage. They may do better than humans in situations where there's movement on all sides.
Sorry, but have you driven in a city in India or Indonesia? Just for example?
My favorite example is a major intersection with no traffic lights. Everyone just drives onto it and then you get a sticky sauce of cars that move at a fraction of the allowed max speed until they leave the intersection again.
Horns and eye contact are used to negotiate progress by all paricipants. I would love to see a "self driving" car on such an intersection.
My guess is a video of this car's encounter would go absultely viral. But not for the reasons proponents of this tech hope for. ;)
Oh you will, soon. Eye contact itself isn't hard, given it's 99% in the mind of the beholder anyway.
You'll see it and then either regret it, or join the cohort of people saying it's not true eye contact because there is no soul behind the window of machine eyes; mimicking the current talk that LLMs are just stochastic parrots without a mind behind them.
If it communicates intent (or even just allows the observer to predict future behaviour - if we're not willing to ascribe "intent" to a robot), it doesn't matter whether it's a plausible soul or not.
At that point we're basically talking autonomous cars with big external-facing emoji screens. "That Fiat is looking red in the face, it's about to cut in".
That's more communication and emotional bandwidth than 99% of driver-to-driver interactions. Someone cuts in and you flip them the finger, how often to they even see it?
