It happens a lot.

Hard problems are rarely just one hard problem, but until you solve the first hard sub-problem you won't even know what the other hard sub-problems are, let alone what proportion of difficulty each contributes.

It's like climbing a mountain in the fog: you've found the way forward and if it holds out you can extrapolate when you will reach the top, only to find you dead end at the face of a sheer cliff...it will be awhile before you find the next way forward.

It's progress but the destination remains further than it seems.

What fundamental problems remain to be solved for self driving cars?

Vehicle detection: detect any vehicle, even from the side, even if its shape is rare, even at dark, etc. This could be solved with data.

Control: when to yield, without watching the face of the other driver, etc.

Edge cases: obeying a police officer, yielding to an ambulance, cooperating with other cars.

And the non-technical side.

Liability: who is liable when the car kills a pedestrian?

Driver engagement: How do you safely transition from automatic control to manual control.

Maintenance: Will the manufacturer be obligated to provide software updates for the life of the vehicle? Even if that vehicle is 20 years old? Even if newer software is 10x less likely to kill a pedestrian?

Cost: Are the costs one-time or will there be a maintenance fee?

Licensing: Does the manufacture have the right to disable functionality after the purchase? Do they have a right to your data? Can they sell that to your insurer?

Regulation: Who certifies systems? How do they test them? When is a system "good enough?"

All of these things sound trivial compared to the technical challenges but it's these kinds of non-technical challenges that killed the small-airplane market in the US and are still unresolved 65 years later.

Nvidia mentioned identifying humans, dogs, parked cars, trucks, street signs, street lights, etc better than a human.

On their todo list was identifying kinda of cars (like a police car), kinds of trucks (school bus and ambulance), and acting appropriately.

I've seen google mention responding to hand gestures from bicyclist. But also being exceedingly polite and repeatedly stopping as a track standing bicyclist rides slightly backwards and forwards.

Not to mention that, since you did robotics in the 90s, there might just be another point where the entire field of AI falls off a cliff, again.

On the one hand, I'm inclined to think not. Having said that, I do wonder if the rapid improvements (and sometimes impressive results) in certain narrow domains are misleading a lot of people to assume that all that's needed is continued incremental improvement.

As a case in point, while voice recognition has indeed gotten quite good with the right kind of microphones (e.g. Amazon Echo), the natural language processing and other "smarts" to use that as a basis for even a subpar by human admin standards digital assistant seems to still be reasonably far away. And that doesn't even require interfacing with the physical world.