self driving vehicle on UK motorways in 2023

[slowsider]

That was my inference

[Horse]

Interesting that Dowd is concerned about Teslas "indiscriminately mow down children."

Would he prefer that they discriminately do it?

[Horse]

That was my inference

Perhaps it was, but - and I haven't seen the videos - until this morning I didn't know it had actually happened. Did you?

[Horse]

https://dawnproject.com/campaigns/

The Dawn Project’s mission is to make computers safe for humanity.


Connecting the power grid, hospitals, and millions of cars to the Internet with software riddled with bugs and security defects has turned these systems into potential weapons of mass destruction at the mercy of hackers.

Well, I wonder how he defines 'safe'?


Dawn Methodology
History
The Dawn Methodology has been developed by our Founder, Dan O’Dowd. Dan has dedicated his life to studying software development. That experience underpins The Dawn Methodology and its supporting technology. Dan has written the software for hundreds of the world’s most safety-critical and security-critical systems. In aerospace and the military these include Boeing’s 787s, Lockheed Martin’s F-35 Fighter Jets, the Boeing B1-B Intercontinental Nuclear Bomber, and NASA’s Orion Crew Exploration Vehicle.
After studying why operating systems contained so many bugs and security defects, Dan developed The Dawn Methodology to design a new operating system from the ground up that could run on any computer system. Its primary requirement is that it must never fail and it must never be hacked. It must also perform the functions required completely accurately and safely.

[slowsider]

That was my inference

Perhaps it was, but - and I haven't seen the videos - until this morning I didn't know it had actually happened. Did you?

Koopman has been all over it on LinkedIn.

[slowsider]

Interesting that Dowd is concerned about Teslas "indiscriminately mow down children."

Would he prefer that they discriminately do it?

I think his point is that they can't discriminate children.

[Count Steer]

Interesting that Dowd is concerned about Teslas "indiscriminately mow down children."

Would he prefer that they discriminately do it?

I think his point is that they can't discriminate children.

From the things that they should mow down?

[Horse]

Interesting that Dowd is concerned about Teslas "indiscriminately mow down children."

Would he prefer that they discriminately do it?

I think his point is that they can't discriminate children.

Actually 'mannequin dressed as child'. The other videos seem to suggest that Teslas can.

But will still drive into random, metre high, objects.

This has echoes of the original, rigged 'elk test'.

[The Spin Doctor]

This time last year...

"Paralympic athlete ruled out of Games after being run over by self-driving bus in village. Toyota has been forced to suspend all self-driving vehicles at Tokyo 2020, after a bus ran over a visually impaired athlete at the Paralympic village, leaving him unable to compete at the Games"

Toyota apologised for the “overconfidence” of a self-driving bus after it ran over a Paralympic judoka in the athletes' village and announced an immediate halt to its all-electric, self-driving e-Palette shuttles—the autonomous buses that had been ferrying athletes around.

And that's in a strictly defined environment. To be fair to the bus, it detected the athlete and stopped, and it was the human operator that restarted the bus, but it then ran down the person. The collision happened despite the crossing actually being manned by two attendants, as well.

You could argue that this shows that the automation worked, but it also shows up the significant difficulty of the interface between human and automation.

Toyota made up their own minds and turned off most of the automated functions.

Toyota Chief Executive Akio Toyoda apologised for the incident on YouTube, saying “a vehicle is stronger than a person, so I was obviously worried about how they were". Toyoda said the accident showed the difficulty for the self-driving vehicle to operate in the special circumstances of the village during the Paralympics with people there who are visually impaired or have other disabilities. “It shows that autonomous vehicles are not yet realistic for normal roads,” he said.

Tesla claims that their cars in self-driving mode are safer than the average driver also turn out to be false - something like 96% of all the mileage covered by Teslas on their autopilot function are on 'controlled access' roads (the only place it's legal to run the car in that mode incidentally) but Tesla compared their mile per crash rate with ordinary cars on ALL roads. Compare like with like data and Teslas turn out to be rather less safe than ordinary drivers on those same controlled access roads.

The wording of the UK announcement is interesting. To my mind, the ‘safety ambition’ of the consultation which has the stated aim of discovering how "this technology can be as safe as a competent and careful human driver" is a telling admission that these vehicles are NOT safer than humans. Autonomous vehicles should be BETTER than a safe and careful driver. Otherwise it's hard to see how the improved safety goals can be achieved.

A standard that only aims to match humans seems to me to be a tacit admission that the performance of a human behind the wheel actually sets a much higher bar than was previously admitted.

[Horse]

This time last year...

"Paralympic athlete ruled out of Games after being run over by self-driving bus in village. ...
And that's in a strictly defined environment.

You'll love this, then:

https://www.cavforth.com/

To be fair to the bus, it detected the athlete and stopped, and it was the human operator that restarted the bus, but it then ran down the person. ... “It shows that autonomous vehicles are not yet realistic for normal roads,”

So human over-rode the AI (and then the athlete).

Bloomin' 'umans, eh? Sooner we get rid of 'em ...

The wording of the UK announcement is interesting.

the ‘safety ambition’ of the consultation which has the stated aim of discovering how "this technology can be as safe as a competent and careful human driver" is a telling admission that these vehicles are NOT safer than humans. Autonomous vehicles should be BETTER than a safe and careful driver.

No-one - AFAIK - has ever claimed that AVs are going to be perfectly, crash-free.

So, to address your concern, what would the crash stats be like if all human drivers were 'competent and careful' all of the time? (Please don't go all Safety Ii at this, 'most of the time it goes right') People are fallible. Machines are fallible. But the machines are capable of continuous improvement in a way that people are not. For most of the time the AI will work however it has been programmed to. And, as you know (if you watched the video I've posted umpteen times) there is the capability to train AIs on crash scenarios in a way just not possible with humans. Does it actually happen? I have no idea

So how can the industry achieve BETTER? By that continuous development. And you know what? It's happening now, right now (well, possibly ) on public UK roads and has been for years. They drive amongst us! Some are AI systems driving themselves, others will be human drivers gathering data on what 'normal' is, others will be simultaneous AI systems running and comparing their decisions with the human who is actually driving.

Here it is again, 15:50 on:

[slowsider]

.

The wording of the UK announcement is interesting. To my mind, the ‘safety ambition’ of the consultation which has the stated aim of discovering how "this technology can be as safe as a competent and careful human driver" is a telling admission that these vehicles are NOT safer than humans. Autonomous vehicles should be BETTER than a safe and careful driver. Otherwise it's hard to see how the improved safety goals can be achieved.

A standard that only aims to match humans seems to me to be a tacit admission that the performance of a human behind the wheel actually sets a much higher bar than was previously admitted.

Leaving aside the issue of bugs in the code , software that only matches a 'safe and careful driver ' will go a long way towards improving road safety. Imagine if all drivers drove like safe and careful drivers: no drunk, no drugs, no fatigue, no tolerated mechanical defects... and if the safe and careful ones never got distracted, or sneezed, or.. you get it.
Humans with experience understand an extraordinary amount about the driving environment that machines have to learn, and humans can comprehend edge cases and adapt strategies to compensate.

[Horse]

Humans with experience understand an extraordinary amount about the driving environment that machines have to learn, and humans can comprehend edge cases and adapt strategies to compensate.

Even doing relatively basic things consistently, like 'not trying to share road space with another road user' might be an improvement.

[slowsider]

Humans with experience understand an extraordinary amount about the driving environment that machines have to learn, and humans can comprehend edge cases and adapt strategies to compensate.

Even doing relatively basic things consistently, like 'not trying to share road space with another road user' might be an improvement.

(Small) kid's stuff

[Horse]

Even doing relatively basic things consistently, like 'not trying to share road space with another road user' might be an improvement.

(Small) kid's stuff

Kids, or mannequins?

[The Spin Doctor]

Imagine if all drivers drove like safe and careful drivers: no drunk, no drugs, no fatigue, no tolerated mechanical defects...

It would make very little difference to the overall safety stats since these outliers actually make up very few of the crash stats. 'High risk' doesn't automatically mean big numbers. Most crashes actually happen to legal, rules-compliant drivers.

and if the safe and careful ones never got distracted, or sneezed, or.. you get it.

The vast majority of 40,000,000 drivers DON'T crash in any given year, in around 300 billion miles of driving.

Interestingly Humans with experience understand an extraordinary amount about the driving environment that machines have to learn, and humans can comprehend edge cases and adapt strategies to compensate.

Exactly. A machine can do no better than the programmer's best guess, and machine learning requires experience too. I suspect that many people have assumed this would be a lot easier to gain than it is actually proving.

[Cousin Jack]

Humans learn from imagination. An overtake, or a bend is suucessfully completed, but the human sees something later, like s huge tracor, and imagines if that had been 60 seconds further up the road where it would have caused a head on collision.

I suspect AI only learns from collisions, or at best, from very near misses.

[Horse]

Humans learn from imagination.

I suspect AI only learns from collisions, or at best, from very near misses.

The consistently repeating three gotcha bike crash types (smidsy, corners, overtaking) suggest that human motorcyclists can't learn others's (or even their own) collisions and near misses.

[Horse]

I suspect AI only learns from collisions, or at best, from very near misses.

FWIW I don't know 'how' they learn. Although I heard someone (who is not directly involved) describe their driving as something like "keep to the rules of the road,"[Highway Code] "stay on the road, and don't hit anything else".

[slowsider]

Humans learn from imagination. An overtake, or a bend is suucessfully completed, but the human sees something later, like s huge tracor, and imagines if that had been 60 seconds further up the road where it would have caused a head on collision.

I suspect AI only learns from collisions, or at best, from very near misses.

Imagine if humans had a wi-fi connection to each other and learned from mistakes made by those linked to them.

[Count Steer]

Humans learn from imagination.

I suspect AI only learns from collisions, or at best, from very near misses.

The consistently repeating three gotcha bike crash types (smidsy, corners, overtaking) suggest that human motorcyclists can't learn others's (or even their own) collisions and near misses.

So the algorithms factor in severity of collision or potential severity of near miss and adjust accordingly. We all know where that leads don't we? Nothing travels at more than 10mph. BRAKE will be ecstatic.