Fully automated vehicles - 'not in our lifetime'?

Arklight

Automated vehicles appear to kind of work in California because there are almost no roundabouts, no cyclists or pedestrians, most junctions are light controlled where no-one advances into the middle unless they are on a protected left turn, and those that aren't use a first in first out priority system. Most drives are multi lane with everyone going in the same direction. Having driven all over LA I can still see an automated vehicle failing at rush hour (which is almost all day) because American drivers simply don't let you out unless you make them.

Driving in the US bears little to no resemblance to the UK, and cars that are tested endlessly in California even would be paralysed in Boston. The only way automated cars are coming to London in our lifetime is if they get rid of the human drivers at the same time.

Our small congested, potholed roads with patchy 4G and worn out road markings, that rely on drivers politely waving eachother out of queues would be incomprehensible to any AI we can develop in the next 50 years.

snowqueen555

You never know, I mean the infrastructure needed for internet and telecommunications sprouted from nowhere.

I would say automated driving will be possible in some situations, 50 years is a long time, and hopefully I will still be alive by then, but looking back 50 years ago, some of the things we have now are the stuff of scifi.

michaels

Humans can actually only cope with limited inputs and processing bandwidth is low so we simplify and pattern match based on experience, ie what learning algorithm neural nets do.

Trying to program an autonomous car to recognise and respond to every situation would be an immense task but Tesla instead just feeds in billions of real miles of sensor inputs and actual human control response and let's the system learn pretty much how 'go' was solved by letting the algorithm play against itself enough times.

Arklight

snowqueen555 wrote: »

You never know, I mean the infrastructure needed for internet and telecommunications sprouted from nowhere.

I would say automated driving will be possible in some situations, 50 years is a long time, and hopefully I will still be alive by then, but looking back 50 years ago, some of the things we have now are the stuff of scifi.

You could perhaps have drive by wire routes with enough external monitoring to allow a robo-car to cope with the human drivers, as a start.

Arklight

snowqueen555 wrote: »

You never know, I mean the infrastructure needed for internet and telecommunications sprouted from nowhere.

I would say automated driving will be possible in some situations, 50 years is a long time, and hopefully I will still be alive by then, but looking back 50 years ago, some of the things we have now are the stuff of scifi.

We can hope so, but being able to imagine something and thinking it should be possible is rather different to it actually existing.

https://qz.com/114699/why-were-a-long-way-from-computers-that-really-work-like-the-human-brain/

The trouble is that at the moment, no computer is powerful enough to run a program simulating the brain. One reason is the brain’s interconnected nature. In computing terms, the brain’s nerve cells, called neurons, are the processors, while synapses, the junctions where neurons meet and transmit information to each other, are analogous to memory. Our brains contain roughly 100 billion neurons; a powerful commercial chip holds billions of transistors. Yet a typical transistor has just three legs, or connections, while a neuron can have up to 10,000 points of connection, and a brain has some 100 trillion synapses. ”There’s no chip technology which can represent this enormous amount of wires,” says Diesmann.

BobQ

GreatApe wrote: »

You also miss the obvious

In aviation the human pilot cost is less than half of one percent of the cost of a ticket
In taxis the human pilot cost is closer to 70% of the cost of the ticket

There is also the cost of the vehicle too. A £50k pilot to potentially protect a £150 million flying machine. A £25k taxi driver only potentially protects a £15k car.

Self drive will first take taxis and HGVs
With taxis so cheap people will very rapidly abandon second cars and then main cars too

Even if self drive taxis were twice as dangerous as humans people would take the robot taxi for £5 rather then the Hunan for £15. But its more likely to be the robot taxi is twice as safe and only a third the price and available much quicker

You may also be missing the point. The saving in "driver costs" cannot be viewed in isolation. It will be accompanied by increased legal and regulatory costs.

The cost of a highly automated plane is more than the pilot, in includes training and pilot assessment, the testing of the technology that enables a plane to fly with such limited pilot intervention and a host of regulatory requirements.

A driven car undergoes a safety check once in a year and the driver once in a lifetime, but how many regulatory checks will a driverless vehicle need to undergo? Will the insurance costs be the same? Will maintenance costs be higher? Who will refuel the vehicles? What happens if the vehicle fails to comply with its specification?

These problems will be solved but there is more to any transport solution than the hourly rate of the driver.

The public accept honest mistakes by car drivers even if they do involve some penalties. But the reliability of driverless cars will take years to achieve and it will only take a few accidents for the public to demand more regulatory checks.

BobQ

GreatApe wrote: »

L
I think we are already there the companies just need a few billion miles of driving to feed their AI self learning programs. Once its out and partially usable they will gather exponentially more data and the AIs will get exponentially better at self driving. Tesla is already doing this, their cars gather human driving and learn from it. In theory they will just get better and better and better. At some point it will be at least twice as safe as humans then turn it on and gather more data and continually improve the AI

AI can indeed learn from humans and get better, but the training required to reach an optimal performance is far from certain and the notion of getting exponentially better is very doubtful.

When the AI exceeds the human capability how will it get twice the capability? By learning from other machines? The real test of AI is whether the trained capability can deal with the unexpected. So for example you can probably train AI to deal with a slow puncture, but can it get the passengers to a place of safety if the tyre blows at speed? Or a drone landing in front of it? Sounds like a fascinating business to be working in!

AKRev01

100% in our lifetime IMO!

MoneyMoney

I heard a car exec talking about this and one of his main concerns was bullying. If you pull out in front of an AI car it will stop 100% of the time, it won't shout or drive behind you, it will politely let you go every time. Human drivers will simply bully the hell out of AI cars, you will be sitting in the back of your AI taxi waiting to be let go all day..

Thrugelmir

As with all these new concepts. The reality might be very different in the future. Perhaps these vehicles will run on specially designed roads in cities and larger urban conurbations. In effect personal taxis. That can be parked/stored automatically when not in use. Then recalled at will.