>
I did not say 80% of the road, but 80% of the travel. Most of the kms driven are on highways, ring roads, and main roads into larger cities.I don't think your experiences are typical for most people the world over. For example in the UK, we don't have that many motorways so most of the commutes by cars are along busy "normal" roads or dual carriageways (which are a little like motorways but with frequent junctions like roundabouts that have complex social interactions and thus might need AI to navigate safely anyway. But even there, most people people in the UK don't spend 80% of their drive on dual carriageways - let alone motorways. Instead we have "main roads" which join towns together.
Commuter traffic in the towns in the Netherlands where I've been were more focused on public transport and bicycles than they are cars - though that's not to say they don't also have a heavy motor-vehicle traffic as well.
I remember when I travelled around America thinking how much different attitudes to transport were to the UK and many European countries. I think that's in part because of how much farther away places are in the US. But I'm now drifting into speculation. My point was your observation about travel behaviours isn't globally representative.
> The car will pull over if you don't take over. I fail to see the issue in that. I'm advocating for a system that will autonomously drive where the infrastructure allows, not everywhere. If the system can handle driving at the highway for me for 2 hours, I'll be significantly more fit for driving the last 20 minutes.
Again, motorway driving is the easy part. It's where there is less to observe, low risk of people stepping out in front of you etc. It's where there are fewer road deaths too. And it's also the easiest part of driving to automate. So motorway driving isn't where we should be concentrating our efforts.
> No you do not. They can even remove a small patch of tarmac, "shoot" and entire sewer line a long way and remove patch of tarmac at the destination. That's how they get optical fibers under roads and brick paving without digging it up.
That's not how they do things here :) But in any case I did say that's still only a small part of the overall cost and orchestration needed in deploying cables.
> the current AI approach will either cost many lives
Will it though? That sounds like baseless speculation to me. Any who's to say your solution wouldn't also have glitches that result in a fatality? Ultimately you're still depending on software to make decisions and even robotic lawn mowers aren't without their own odd hiccup.
> Again, a 100% solution is never happening.
I know. That was my point. It's not going to happen and partial coverage isn't good enough.
> But by legislation we can increase adoption slowly, but steadily, until and past the point where we see the many benefits.
50% coverage isn't enough. 90% coverage isn't even enough. 99% coverage isn't even enough but at least you're starting to get close. And how long will that take? Lots of people are still driving 20 year old cars every day. So lets say 10 years to design, develop. 5 years to then standardise the specification. 5 years of bureaucracy to get legislated then another 5 years for the legislation to become effective (because governments like to give a warning). They you need to wait 20 to 30 years until most people have bought a new car. That's 55 years at a minimum and you'd still not cover the thousands of people who like to drive classic cars. So you might as well double that figure. By which point AI has long since caught up with the ability to sense other cars on the road (because we've already mostly solved that specific part of the problem anyway) and at a fraction of the cost.
I've already said that if we were to build infrastructure from scratch then your approach is definitely the way we should progress. But with the way our road infrastructures run at the moment: the vehicles on there, the hazards that can occur, the cost of maintaining the infrastructure and how that's even funded; well your solutions are simple not feasible for the real world.
AI is one of those occasions where worse is better. I completely agree with you that your solutions are better - technically speaking - at solving specific problems related to driverless cars but it doesn't solve all the problems, it's more expensive to implement, can't be deployed everywhere, and yet it will still take longer to roll out than our current tact of developing AI. When you actually look at each problem in detail your proposals fall far short of being workable.
These things aren't built without considerable thought put in them and I seem to recall that early prototypes of driverless cars did work on the principle of having smart cars that communicated with each other and smart roads too. Essentially everything you've been advocating. So it's not like you're suggesting anything that hasn't already been tried. But every lecture and documentary I've watched on this subject come back to the same final conclusion: for self-driving cars to become a thing they need to build the car to drive like a human would because changing the roads and the cars on it is actually a harder problem to solve.