I liked the article, but I did wonder if the author understood what self-driving vehicles would allow for, providing that they had maybe 7-8 nines in reliabilit

I liked the article, but I did wonder if the author understood what self-driving vehicles would allow for, providing that they had maybe 7-8 nines in reliability and safety (if that is even possible). People have already mentioned the idea of "sending the car back home to park" and "being able to do something else in the car while commuting". But there are other things the cars could do: 1. Trains of cars - convoys - drafting each other, reducing energy use, while also increasing road density (bumper-to-bumper traffic - but at highway speeds). 2. No more "traffic waves" on the freeway, nor "rubber necking" at accidents, both of which likely contribute to the congestion issue. 3. No-stop intersections - again at high-speeds - because cars could coordinate/negotiate their speeds and lanes to allow for "collision-free" operation (unfortunately, there will still probably be collisions - and they wouldn't be pretty). All of these options - and possibly others I haven't thought of - rely on things that self-driving vehicles don't yet have - mainly car-to-car communication - so all cars know where other cars are (even if they aren't within sight-lines of the sensors) and how fast they are moving. Basically a form of swarm robotics, with constraints on motion vectors. Of course, #3 (and maybe the other two) might not come about unless the chance of an accident drops to infinitesimally small values - perhaps it may not even be possible. Better safety systems for the occupants would likely need to be developed as well (for that "impossible" chance it would be needed). Such cars might also need to be "trained" on "how to crash to protect occupants" and/or how to coordinate a collision with another vehicle(s) in real-time as the collision unfolds, to minimize occupant danger - this again, would require a standardized mesh network and coding to be able to take advantage of all the sensor data that would flood in (and out - as sensors and actuators become disabled by the accident unfolding - maybe the other cars involved - or not involved - could share their information "from the outside" to allow the disabled car to take better actions or whatnot). It's actually a fascinating problem when I think about it - extremely complex, lots of parameters and such; I wonder if anyone is working on it? If not - they should...