Fully automated vehicles - 'not in our lifetime'?

kabayiri

Herzlos wrote: »

...
You're taking data from dozens of sensors, processing them all individually and using the complete picture to determine what to do with the car.
...

Have you seen how under-utilized one of the chips is in your smartphone?

I'm curious, where is your experience of scene analysis from?

westernpromise

Herzlos wrote: »

A modern smartphone just doesn't have the power to drive a car. Honestly. Look it up.


You seem to think it's a case of "if a camera can record it, we can drive" but you need to do a huge amount of processing on that visual data - object detection, speed tracking, spacial identification, for dozens if not hundreds of objects at once. You need to identify from visual queues where the road is going, what's likely to intersect you and if any deviations are required. You need to have some awareness about what's around you in all directions.



You're taking data from dozens of sensors, processing them all individually and using the complete picture to determine what to do with the car.



Your phone can probably read all the road signs in front of you but that's not enough.

The gadget in my car that claims to identify speed limits works by reading the signs and repeating them on the dashboard display as you pass them. I can think of 2 ways of doing that - one would be to infer the angle of tilt by how circular or ovoid they appear, and then using trigonometry to work out when you have drawn level with them. The other way would be to measure their apparent size at intervals and work out from the change when you will reach the sign.

There may be other ways and they all require rather clever processing, but as clever as it is, my car still routinely mistakes the black and white chevron signs on roundabouts
https://www.drivingtesttips.biz/uk-road-traffic-warning-signs.html
for the national speed limit sign
http://drivingconnections.com/uk-road-signs/
and tells me the limit is 60mph when I'm actually still in a 20 and approaching a hazard!

There is a function I've never worked out how to use whereby the cruise control sets itself to the most recent speed limit sign. Given the above propensity to error it seems like it would be horribly dangerous to use. Maybe in another 20 years?

kabayiri

This is a good book for those interested in digital scene analysis

https://books.google.co.uk/books?id=4Yz3gLkISnwC&pg=PA25&lpg=PA25&dq=agin+digital+imaging&source=bl&ots=twiCGyCTNP&sig=qRyyjv6wVh6zGzLardupIS4AhjQ&hl=en&sa=X&ved=0ahUKEwiw8Zv1m4_bAhWGCcAKHR32DqUQ6AEIQDAE#v=onepage&q=agin%20digital%20imaging&f=false

It's probably quite out of date now though.

The bit linked to mentions medical imaging. The fundamentals of that were done by mathematicians 150 years ago ! Amazing to think.

Herzlos

kabayiri wrote: »

Have you seen how under-utilized one of the chips is in your smartphone?

I'm curious, where is your experience of scene analysis from?

Have you seen what your phone is doing when it's under-utilized? Or how slowly things like facial recognition works? Try doing facial recognition on 8 video streams in real time whilst your gps app is running (not suspended).

My job tangentially involves industrial imaging from time to time. It's not my speciality but I'm aware of it.

kabayiri

Herzlos wrote: »

...
My job tangentially involves industrial imaging from time to time. It's not my speciality but I'm aware of it.

I worked on graph node reduction for a while, which is why stuff like TWEAK caught my eye.

Thing is, it must have moved on since I reckon.

When you flip the image space from cartesian to MC space, it changes the way you look at lines and intersections.

Dedicated scene processors and parallel processing offer an ability to distribute the load. Creating a cooperative network....that's where the challenge might come in.

(oops, just for Hamish's benefit...I have to say this : robots, robots, robot, lol)

HAMISH_MCTAVISH

kabayiri wrote: »

(oops, just for Hamish's benefit...I have to say this : robots, robots, robot, lol)

Aye... most cars will be truly driverless... in 50 years or so.

Maybe another 50 after that before bipedal humanoid robots can actually do many/most human jobs autonomously.

Guess we'd better import some labour until then....:p

kabayiri

HAMISH_MCTAVISH wrote: »

Aye... most cars will be truly driverless... in 50 years or so.
...

I'm not that interested in the autonomous car, personally.

I can see gains in an intelligent road network. Our newer car can already park itself, and is always connected to live feed information. The fundamentals are there.

If cars could just operate as a collective during key peak windows on major motorway or highway hotspots, we could squeeze out extra capacity.

The popular 'me me me game the system' trick around here is to use the left hand lane, labelled up as leading off the Motorway, to jump the queue on the right hand lanes, and then these people start indicating to get back in...blocking both the main through route and the off route. Ultimately, they seek to gain personally, but they slow down progress for the whole.

westernpromise

ilovehouses wrote: »

So it's done by reducing proximity risk by having lower maximum speeds or bigger distances between vehicles. Nobody worries that the lives of Stansted shuttle train users might be a lower priority than those on the Singapore metro. That's an extreme obviously but the software just doesn't need to worry about who is more important.

I doubt the technology is going to be sufficiently advanced anytime soon that we need to be concerned about Tony Blair having a God version of the software.

That's not how it would work though. Tony Blair and the elite's version would create that bubble of safety around him by moving other people's cars out of the way.

Otherwise every time a car cut in just in front of the orange war criminal's car, he would have to slow down to maintain a safe distance.

Think of how the Soviet class system worked - it will be like that. There will be certain traffic lanes reserved for the nomenklatura; speed limits will applicable to and observed by only the proles' version of the software, but not to Tony Blair's cars, etc. When Tony Blair wants to go through traffic lights, all other cars will be be stopped by his software until the Blair motorcade taking him to the restaurant or whatever has passed.

And of course the software will expend other people ahead of Saint Tony.

Essentially it will be a version of the software fitted to emergency vehicles.

GreatApe

Herzlos wrote: »

A modern smartphone just doesn't have the power to drive a car. Honestly. Look it up.

You seem to think it's a case of "if a camera can record it, we can drive" but you need to do a huge amount of processing on that visual data - object detection, speed tracking, spacial identification, for dozens if not hundreds of objects at once. You need to identify from visual queues where the road is going, what's likely to intersect you and if any deviations are required. You need to have some awareness about what's around you in all directions.

You're taking data from dozens of sensors, processing them all individually and using the complete picture to determine what to do with the car.

Your phone can probably read all the road signs in front of you but that's not enough.

Assuming you had a dedicated very high bandwidth feed do you not think you could drive a car remotely with the 1080p feed off a smartphone camera?

If the answer is yes then it has all the hardware you need, it doesnt have the software in this case the software is your brain but the hardware is in place.

Herzlos

GreatApe wrote: »

Assuming you had a dedicated very high bandwidth feed do you not think you could drive a car remotely with the 1080p feed off a smartphone camera?

If the answer is yes then it has all the hardware you need, it doesnt have the software in this case the software is your brain but the hardware is in place.

Only if you don't understand how it works.

You could sort of drive a car with a video feed bit you'd lose peripheral vision and depth perception. The lag would be fatal too.

Getting the sensor information onto the computer is the trivial bit; just make sure you have enough pins. Doing something useful with it is the hard part, and that currently requires much more processing than a phone or even a high end desktop can do. They can do bits in isolation but you need the full picture to drive