Imagine the implications.
We have had fully autonomous airplanes and trains for decades now. The technology has been shelved because of legal reasons, not because it doesn't work.
Self-driving technology will only last until the first child gets run over by a self-driving car, at which point the technology will be banned. Lawyers will need a fall guy to take the blame, and since tech bro corporations have a EULA that absolves them of all responsibility that will be the end of self-driving.
The problem with self driving technology is that companies like Tesla want to keep you liable for the decisions their AI makes. That's why you have to sit behind the wheel and still keep an eye out.
Here in the Netherlands there were several tests with self driving trains, but the systems didn't prove satisfactory yet. The systems are being worked on, though. Putting passengers in the hands of automated systems requires a lot of care and responsibility, and a few companies are willing to take that risk (for plenty of money, of course).
As for planes, autopilot is used very often. Automated landing is mandatory in some runways to keep time between planes down, though pilots don't seem to like the aggressive touchdown these systems provide. We'll never have fully autonomous planes because someone needs to be around when the plane runs into trouble; computers won't land you on the Hudson.
In certain situations humans do not respond in an ideal way. They are bad at deadling with conflicting information. And will often make poor choices when faced with an emergency. A suitably designed automation may well do better in those situations. And a lot of accidents now are exactly those kind of situations. They are survivable incidents where the pilot fails to take the correct action.
I don't think we should give emergency control to autopilots just yet. But the scientific problem it poses is worth tackling. How do you deal with possibly conflicting information and be sure that the reported situation and plan is correct. That is the problem that underlies a lot of challenges (like ATC, IFR navigation, search and rescue, routing drones, even command during a war). You have a set of data from the machine, recorded data about the external world, data from other machines and control, rules about expected behaviour, mapping data, model outputs, weather information. And that all has to be tested, fused together and then acted upon.
On the surface, yes. Looking more closely, kind-of-but-not-really.
A plane auto-pilot is not really meant to fly the plane from gate to gate. It's more a series of automation simplifying each action performed during the flight. But each action have to be triggered individually and controlled by the pilots.
For trains, full automation is not common at all. You see it see it mostly on a few metro lines, which is an ideal case (platforms can be secured, homogeneous equipment, very static route, short track enclosed in tunnels). On a classic railway networks, we are still at the stage of a few experimental projects.
So, even with trains and planes, we are hardly there yet. And these are much simpler systems to automate: they are operating in heavily controlled environments, with a traffic that is not that dense (compared to a city street) and with tons of help from the infrastructure (think ILS for plane landings, or ETCS signaling & speed control for trains).
So, a self-driving car in an open environment like a street? No, we are there yet, and in my opinion, we are still one or two decades away from seeing that.
> exceptional circumstance appears
> no network
> implications