I still don't understand why the "autonomous driving" people aren't concentrating on freeway driving--possibly with special roadway markings--over local driving.

I don't really mind driving the 10 minutes to the store.

I mind driving 90 minutes between San Diego and LA. I mind driving 90 minutes between San Jose and San Francisco. I mind driving 7 hours between LA and SF. I mind driving driving 24 hours (two 12 hour shifts) between San Diego and Austin.

And for those of you suggesting car rentals, have you seen the price of car rentals even before Covid? It was getting to the point that car rental was often more expensive than airfare.

> special roadway markings

I continue to expect something of this nature to be necessary before we get self-driving tech anywhere close to where we want it to be. Cameras simply don't work in all conditions, and even with LIDAR mistakes can be made. There needs to be an external reference point.

But currently eyes do. So theoretically cameras should right?

Lol - if it were that simple we would have had this tech years ago.

The physics of the eye are relatively complex, the data processing and recognition we do more so, and the ability to dynamically mentally model a set of controls onto a vehicle of arbitrary shape and dimension even more so.

Cameras are not eyes, the algorithms we use to detect are completely different, the algorithms we use are completely different.

I can't prove that it is impossible, but its different enough that it might be. A bit like saying a tomato and a brocoli are both vegetables with organic matter, but good luck making Tomato Juice with broccoli.

I think most of the car manufacturers are starting to roll out automatic lane centering for the highway in the most recent models of cars.

Also see Comma's Openpilot for adding automatic lane centering to a recent 3-4 year old car from Honda, Hyundai, Toyota and others.

If we could have true autonomous local driving, cities would change forever. All cars would live outside the city, and cover (pooled!) demand with incredible efficiency.

Cars in cities today are incredibly stupid compared to public transit. Full autonomy could change that.

Ah yes, artificial artificial intelligence.

Ask yes, exploited human intelligence.

Though also, at the same time the liability questions would quickly become... Complicated.... With that model

In the same way that basically the entire modern global economy is exploiting humans? Well it does, but it also gives something back: living together in a civilization that kinda works (so far). I think for a lot of people doing some mechanical turk style driving could actually be quite a satisfying job, more so than e.g. data entry tasks.

Amazon's MTurk still exists!

"Beetle - why did you just ram us into a wall?"

'I'm sorry! Hit a lag spike, blame Telus.'

Driving along the freeway at 100mph and suddenly "sorry afk bro need to go to the toilet"

Easier solved than with a real driver, just flip a switch and transfer control to another drone pilot. Solves more issues: boring long trip, flip drivers every 30 minutes. One moment you're driving a car in texas, next moment you're driving a car in new york, and 30 minutes later you're driving the Hana Highway on Maui.

With no consequence to you if you decide to go GTA on the system, no peripheral vision, no feel of the road or wheel, assuming you at least have stereoscopic vision with a headset it’s still not at all going to work for white out or hard rain.

It’s a fun idea that we can create a whole generation of Bangladeshi click farmers that drive people instead… but it’s a nonstarter for a bunch of reasons.

Why would there be no consequences? The person would have a driver contract, have passed driving tests, etc.

Not sure what you mean with Bangladeshi click farmers

This is how a lot of ideas get nonsensical holes put in them.

There would be licenses, regulations, threat of lawsuit, qualifications and all sorts of other mandatory requirements. If someone wants to be cynical then, yes, a lot of these safeguards would not be put in place until after a certain number of deaths had been exceeded. eg seat belts, air bags, traffic lights, speed limits, road rules in general.

The idea is still viable.

> There would be licenses, regulations, threat of lawsuit, qualifications and all sorts of other mandatory requirements

These already exist in Taxis, and have you ever been in a Taxi? Their drivers aren't exactly the pinnacle of driving excellence.

Sideline: face recognition based assassination.

I am in a country that is far away and experience multi-second connection drops on a regular basis.

Highways are simple enough that car AI already works fine.

Human intelligence could be useful for navigating construction work.

Really, what we need is to overhaul road construction (and maybe roads in general) signage and protocols to make them machine-readable, not just human readable.

Like in Neal Stephenson's Diamond Age. Poorly paid actors in boxes acting as voice assistants.

Similar to what the company Spinvox used to do. Artificial Artificial Intelligence, they would break up audio recordings and have different people transcribe them, then piece the transcriptions together. They sold the service to wireless companies so customers could get automatic voicemail transcription.

If cars driving themselves will asymptotically approach say 90%, 95% or 99% or situations it can handle, then the car that also handles the last few situations will be very attractive. That's the car that can drive me home from the bar, or that a blind person can own. The 90-99% cars are still for able bodied people with licenses.

At some price point, it will always be a lot cheaper to add humans than it is to make the cars "self driving". And the last % I'm almost certain will never be cheap enough for autonomy.

There's a large and perhaps unsolvable problem of the state of the art solutions that doesn't get nearly as much attention as it should - even if the car can safely handle 99% or 99.9% of the situations, the algorithms are not really able to detect if they're in the last 0.1%. The car will slam into stationary trucks, suddenly brake for no reason, consider the tollbooth to be a bus - all with utmost confidence it's doing the right thing.

In other words - the value of proposition of autonomous driving would be completely fine if the car gives up on unarguably difficult situations and alerts the driver to take over. As of now, however, the car will boldly do something stupid and dangerous.

That "brake for no reason" is a big one. I have a 2015 GMC SUV that has "adaptive cruise control" which I (used to) love.

Since my last service, it seems they've updated it and it's far more sensitive than it used to and seems to break in many situations where it previously did not. What sucks about that is the vehicles behind me are caught off-guard with no obvious signs of the need to break and then the jackass in the GMC SUV slams on his breaks.

I had a similar issue, turned out the tech had moved the sensor and it needed recalibration

There are some models that provide the confidence that they have in the area of data being explored. I forget the name, but I think it's Bayesian Deep Learning. Something like: if it's something they've never seen, they would know they can't take a choice with confidence.

If the car alerts the driver very seldomly, like once or twice in the lifetime of the car, then it could be a dangerous situation as the driver would be somewhat shocked. That or the driver would need to struggle to remain vigiliant as, day after day, the car did the right thing.

In some countries you could even get a real driver (sitting inside your car) for that money...

That was literally my first idea after reading the title. I hope this plan meets a sad ending.

I also see people very quickly stop using their autopilot, as you will think ca. twice a day: "so do I really want to spend 8.50 right now?"

A lot of the early 'voice recognition' did this; the audio went real-time to far-away countries where a human would press the appropriate buttons.

Makes me wonder what those humans are now doing for a living.

Giving that Waymo video about a couple of weeks ago, that seems to be a very poor experience right now (even with the drivers close by).

latency might be an issue though

The average reaction speed of a human to unanticipated situations can be ~3 seconds. Any connection within 1000KMs is generally going to be under 0.1s - an order of magnitude less than a human's.

I'll take your word for that, but remain unconvinced. I don't think it's just about reaction time.

Playing a game with input lag is terrible because you always need need to do all your inputs ahead of time in anticipation what a frame 100ms in the future will tell you. That might make precise control very difficult, not just in unanticipated situations such as e.g. emergency braking.

Games generally already involve adrenaline and focus, where you're primed for action to happen.

Unexpected road accidents typically involve someone who may not even be thinking about driving, who then have to notice the random screech, focus their eyes and process what's happening, then determine they need to immediately break, then break.

The problem with input lag isn't that human input is under 100ms, it's that too much of a gulf between input and on-screen response breaks the feedback loop - your brain struggles to connect your input alterations to an alteration actually happening. If you're throwing a ball, when you twist your wrist the ball is twisted immediately when the wrist holding it is twisted (because it's physically pinned to the hand that's physically attached to the wrist, and literally cannot be out of sync with the fingers gripping it).

It's like how VR users need under 30ms of latency or they throw up, except less drastic.

You can play high res FPS games on remote GPUs now, so I expect that part could be managed, if not now then in the very near future.

Presumably those remote gpus are not halfway around the world. Speed of light sets an inherent floor on the latency.

The issue is not the idea of remote control per se, though connection drops etc. would be a risk, but the notion yoga can outsource it to somewhere with far cheaper labour.

Ah, yes, that's a good point.

How big is the different between simracing with iracing for example, and what would be required for real remote driving?

How is this solved for military drone pilots that do precision bombings in far away countries?

With multiple pilots that manage different aspects of the mission. Also, up in the air at cruising altitude the latency is not as critical.

Not to mention connectivity

As long as connectivity was 'good enough', the car could have just enough self-driving capability to safely pull over in the case of an interruption. Would still need to be some nines, but wouldn't need to be utterly perfect.

I don't think safely pulling over a car is any easier to automate than the rest of driving. Those are some of the most complex maneuvers where you're at least as likely to hit a civilian or another car as driving down the street.

Frankly I am amazed we don't already have remotely piloted vehicles.

Sadly we do, they're just called drones and they are used to kill people. (Sorry, don't mean to be so dystopian)

Would probably do a considerably better job than actual autonomous driving for the foreseeable future too.

There's a future for all these gaming addicted youths after all.

Yes, just hide an Indian in the trunk...