Comment by Creationer | Hacker News Reader

Comment by Creationer | Hacker News Reader

We might have to contemplate that we need to change our roads if we want automomous cars.

We have lanes for carpooling - why not lanes for autonomous vehicles? The road surface could have special markers to centre the vehicle, and to position it in the correct lane for a certain destination. This could be a single lane in a highway, with shorter distances between cars (relying on the quicker reaction speeds of computers and the modern, well-maintained brakes of the cars themselves), allowing humans to overtake and recognise them easily.

Then we need a standard method of communication between vehicles - braking, movement etc. - rather than relying on cameras and radar.

Remove as many variables as possible and the task becomes much easier.

Heck, we'll need a new method of charging road users for road use, without gas taxes, so we might as well introduce a standardised road pricing scheme as well (that takes into account vehicle size, weight, time, and location to calculate the price/km - and can communicate that to the end-user to allow them to optimise their travel).

Driving on highways is relatively trivial with current tech (with a bit of exaggeration for effect) compared to surface streets which don't lend themselves to these neat type of solutions. This is the last mile problem once again and as always that's where the majority of difficulty and cost lay.

This could work well for commercial transit - have standby drivers for last mile, and have them board at the off-ramp and disembark at the on-ramp. For personal transport this could be more difficult though.

Does the last mile problem really need to be solved to make autonomous vehicles fantastically useful?

A vehicle with an autopilot that's as reliably safe a highway driver as I am would be awesome. It'd give me back hours of attention (or sleep) during commutes or interstate travel. Even if it would seek a parked or human-piloted state on encountering problematic weather.

All-weather door-to-door capability may be what's needed to replace a large chunk of individually owned vehicles with a shared on-demand fleets, but highway-only would make individually owned vehicles much more convenient.

I agree, and I don't understand why this concept isn't given more consideration. If, as other commenters have said, autonomous freeway travel is so "trivial" why can't we reap the benefits now while the industry grinds away at edge cases for another 20 years?

You can't rely on what other cars are telling you because a bad actor could create a device that sends out false information that leads to crashes (send out "I'm now accelerating from 50mph to 60mph" then brake immediately, or even just not accelerate and continue to go 50mph). You could of course solve this by verifying all the information sent by cars, but then you have negated any benefits because you have to be able to detect anything they could send in a message.

I think the potential for 'bad actors' on the roads is significantly reduced just by eliminating the human driver!

How come? As it stands, anyone can be a bad actor but their impact is limited, and it requires a pretty big sacrifice (their health). For example, nothing is stopping someone from driving the wrong way on the highway other than a couple of Do Not Enter signs, but by crashing their car into oncoming traffic they risk losing just as much as the people they crash into.

We have a justice system that would prevent this. Same reason why I don't worry that a gunman is going to kill me every time I go out into public.

This argument doesn’t hold though because a single bad actor who figures out how to spoof the interface has tremendous leverage to hurt a lot of people at once. And it’s likely they could leave a device at night and trigger it days or weeks later. One instance of this could literally cause enough public fear to get AVs banned for years.

I agree with OP that each car managing its own cameras and radar makes the most sense. Cooperation by way of individual optimization is the cornerstone of modern society and something we should absolutely imbue our autonomous vehicles with.

>And it’s likely they could leave a device at night and trigger it days or weeks later.

same could be said about leaving a bomb on a bus, and yet there aren't bombs going off on busses every day, and even when it does happen, it's not enough to scare people away from busses for good.

Society kinda works because the vast majority of people are not homicidal.

When you think about it, it's amazing there isn't more terrorism.

This is true. Not the sentence I expected to give me hope on a Sunday night, but it did.

True, and that is certainly not due to impotent efforts to increase security in basically every facet of life.

Bus bombings are (relatively) common in some countries. One went off very recently, actually: https://www.theguardian.com/world/2019/may/19/egypt-roadside...

" and yet there aren't bombs going off on busses every day, "

I don't know if that's true (except maybe in the literal sense that it's not every day, just frequently). As far as I'm aware this is a pretty big problem.

"Society kinda works because the vast majority of people are not homicidal."

I agree with this, and that self driving cars will only work under this assumption, but I don't think having vehicles communicate for safety-related (braking) information is the solution at all. If you're relying on that and can't get by with observed data alone, you open yourself up to a huge accident caused by a tiny rock or bird or anything unexpected (way more so than we do now).

If it's just supplementary as an extra layer of safety (attach this pod to your bike to make double sure all cars see you, but they're also still looking) then sure.

Yup, that justice system seems to stop all those gunmen shooting people.

Except for, you know. The occasional bad actors.

How would one do this? Wouldn't all cars in public road have software signed by strong cryptography?

Cryptography just guarantees that the message hasn't been tampered with or read by a 3rd party, there's no way to mathmatically prove that you will accelerate

> why not lanes for autonomous vehicles?

That would be such a huge change to infrastructure that "on-demand-trains" probably would be the cheaper alternative. There are also a lot of cases where you just lack the space for additional lanes.

Road are so expensive as they are, you want to ask for even more spending on roads?

Roads aren’t actually that expensive. Resurfacing a highway costs about a million dollars a mile. Rebuilding the track on the CTA Redline recently cost $40 million per mile, and that was a miraculously cheap and fast rail project. A new freeway costs 5-10 million per mile. A new subway tunnel can cost a billion dollars per mile. Building the DC Silver Line through existing freeway medians cost $240 million per mile.

Maybe in the boonies. Adding a carpool lane on 10 miles of the 405 cost 1.1 billion and construction was a shit show. Then we need to add sensors and whatever else into the road surface so private companies can get their profit margins? We can't even maintain our existing road surfaces. We have the lanes already, remove a car lane on every road over 1 lane wide and make it bus only; same grade separating benefits of a subway but requiring zero investment beyond retiming lights and buying plastic bollards and maybe a few more busses to pick up demand. The insurmountable difficulty is in not letting emotional NIMBYs kill the bill.

https://xtown.la/2019/05/02/the-sepulveda-passs-failing-grad...

Some UK costs for adding a lane for comparison - https://publications.parliament.uk/pa/cm201617/cmselect/cmtr...

(roughly £4.5-6.5M a mile, probably $7-10.5M in 2015 money?)

Reply on news.ycombinator.com