Also humans can learn from this too. It’s not hard to imagine a new type of safety cone that has an explicit sign for autonomous vehicles to stop far far away.
Two things:
The promise of self-driving cars is that they can operate themselves in an environment not specifically designed for them. If that promise is broken, and it's the position of the industry that things like traffic signals, construction, emergency services, road markings etc need to be modified for functionality with self-driving cars then I think public will to accommodate them dries up quickly.
That being said, I can imagine having a lot of fun with a physical token that can remotely stop an autonomous car.
Car companies will never build in failsafes like that, it would be too easy for anyone to put up those signs anywhere. Cars and streets are made for human drivers, there is no way around it and no alternative to making your car AI as good as human drivers. If you can't do that, you can make a much simpler train "AI" and put it on tracks. We did that in the 60s.
Isn't that kind of defeating the point of autonomous vehicles if we need to build a world that will accommodate them?
It'd just be done such that new signage is made to be easily readable, so that over time, as old stuff is replaced, the signage becomes more readable for machines.
This sort of thing can also benefit regular vehicles.
I'm not sure what "This sort of thing can also benefit regular vehicles." means in this instance. If the cone is machine readable how does that help me?
The cones being machine readable would allow regular vehicles to also be given the ability to read them more reliably (and potentially with less equipment than needed by a current autonomous vehicle), so, they could do things like warn the driver, which could be valuable in reducing the cases of distracted drivers slamming into closed off lanes.
This comparison doesn't make much sense. We're highlighting a dangerous failure in capability here, not a desire to change out the engine. Either engine works perfectly fine, it's secondary properties that we care about with transition. It's a primary responsibility of an autonomous vehicle to be able to identify and react to obstacles.
We either need the enhanced cones or we don't, but only a fraction of cones being enhanced means they aren't actually that useful or that the situations without them are inherently more dangerous. Are we OK with a situation where autonomous vehicles identify the enhanced cones 90% of the time but the non-enhanced 50%? No, we want both to be very high.
I agree that we should strive to improve the vehicles' ability to detect even unenhanced cones. Doesn't mean that we can't also aim to gradually improve detection accuracy further with enhanced cones. That is, if autonomous vehicles identify regular cones correctly 98% of the time, but enhanced cones can be identified 99.99% of the time, we should be okay with a gradual transition to the latter, even though tautologically it means that the former is less safe.
We already do this with all sorts of things, safety standards improvements often have at least a grace period during which the comparatively less safe things are still allowed to exist and operate alongside the safer ones.
I dont believe thay every car maker will not try to invent their own models