self driving vehicle on UK motorways in 2023

[exportman]

Automated Vehicles (AV) Bill

This bill introduced into Parliament on November 8th is the governments most recent step to legislate for self driving vehicles.
The government release describe this as “one of the most comprehensive legal frameworks of its kind anywhere in the world for Self driving vehicles, with safety at its core”.

It goes on to say “Self-driving vehicles could help reduce deaths and injuries from drink driving, speeding and driver tiredness, with 88% of road collisions having human error as a contributory factor”.
One must ask the question if 88% of accidents are human error why have successive governments only focused on speeding?
In the release it states “Every authorised self-driving vehicle will have a corresponding Authorised Self-Driving Entity– often the manufacturer – which will be responsible for the behaviour of the vehicle when self-driving”
As already seen the present crop of driver assisted vehicles have serious short comings especially when encountering Cycles and Motorcycles on the highways.
It is hoped that the legislation will protect these other road users from harm, both physical and financial, as it is certain the corporations producing these vehicles will have expert legal teams to defend them should anything untoward happen.
Read the full release here New laws to safely roll out self-driving vehicles across British roads - GOV.UK (www.gov.uk)
https://www.gov.uk/government/news/new- ... tish-roads?

[Horse]

It goes on to say “Self-driving vehicles could help reduce deaths and injuries from drink driving, speeding and driver tiredness, with 88% of road collisions having human error as a contributory factor”.
One must ask the question if 88% of accidents are human error why have successive governments only focused on speeding?

Apart from drink-driving laws, introducing drug wipe tests, introducing the hazard perception test ... plus manufacturers introducing driver monitoring to identify drowsiness.

[wheelnut]

Never been on motorways in the 'rush' hour?

Cue lots of BMWs holding trucks up in lane 1 at 37mph with the drivers glued to Game of Thrones while not noticing traffic has sped back up.

I appreciate they’re still in a developmental phase but autonomous cars can’t be half measures.

In fact, this sounds a lot less sophisticated than teslas autopilot.

[Horse]

In fact, this sounds a lot less sophisticated than teslas autopilot.

I'm not - never really have been - heavily involved with 'road car' AV development or testing.

But I don't know how Tesla get away with it. I'm fairly sure that AV stuff allowed on-road in the US wouldn't pass UK safety case testing.

[The Spin Doctor]

Meanwhile General Motors have just pulled their fleet of Cruise fully autonomous robo-taxis off the road after California withdrew their licence. That was following an incident where one of them ran over a woman who'd been previously knocked down by a hit and run driver. The woman was dragged under the vehicle as it pulled over to the side of the road and stopped.

[The Spin Doctor]

Are there many structurally separated carriageways that have a speed limit of less than 40mph?

M25 most of the time, most of the way round... it may not be a speed LIMIT but it's frequently the speed of travel.

[Dodgy69]

I'd of thought nearly all accidents were caused by human error, that will just swap to AV error.

[Mr. Dazzle]

But I don't know how Tesla get away with it.

They don't, in a nutshell. Hence why Merc and BM make a show of their homologated status.

[Taff]

I think I wrote this before - as a safety engineer, I really don't know how anyone could write a convincing safety argument for a self driving / autonomous vehicle operating out in the wild.

And once the argument had been written who is going to be able to review and agree it stacks up?

The use of AI, machine learning doesn't fit into the current methodology of demonstrating safety goals have been identified and met, and the industry developing the tech is miles ahead of the regulator, so who is going to write the regulations if the regulator doesn't understand it well enough?

[Dodgy69]

I'd imagine putting your head in a crocodiles jaws is frightening, but sat in one of these doing a crossword is on another level.

[Mr. Dazzle]

I think I wrote this before - as a safety engineer, I really don't know how anyone could write a convincing safety argument for a self driving / autonomous vehicle operating out in the wild.

And once the argument had been written who is going to be able to review and agree it stacks up?

The use of AI, machine learning doesn't fit into the current methodology of demonstrating safety goals have been identified and met, and the industry developing the tech is miles ahead of the regulator, so who is going to write the regulations if the regulator doesn't understand it well enough?

What's the safety case for letting a human drive though?

Passing the driver's test? So....what if an autonomous vehicle passes that?

People have basically been grandfathered in haven't they?

[Mr. Dazzle]

It goes on to say “Self-driving vehicles could help reduce deaths and injuries from drink driving, speeding and driver tiredness, with 88% of road collisions having human error as a contributory factor”.
One must ask the question if 88% of accidents are human error why have successive governments only focused on speeding?

Apart from drink-driving laws, introducing drug wipe tests, introducing the hazard perception test ... plus manufacturers introducing driver monitoring to identify drowsiness.

And the fact when you do crash speeding pretty much invariably makes the results worse.

[Taff]

I think I wrote this before - as a safety engineer, I really don't know how anyone could write a convincing safety argument for a self driving / autonomous vehicle operating out in the wild.

And once the argument had been written who is going to be able to review and agree it stacks up?

The use of AI, machine learning doesn't fit into the current methodology of demonstrating safety goals have been identified and met, and the industry developing the tech is miles ahead of the regulator, so who is going to write the regulations if the regulator doesn't understand it well enough?

What's the safety case for letting a human drive though?

Passing the driver's test? So....what if an autonomous vehicle passes that?

People have basically been grandfathered in haven't they?

That's two entirely different cases.

The safety case is all about the machine, demonstrating that you have a complete understanding of how it can fail / go wrong and the consequence of the failure, and arguing that you've done everything practicable to reduce the risks to an acceptable level. Doing this for mechanical stuff is relatively straightforward
Demonstrating software is error free is a lot more difficult, doing this for AI / machine learning I'd argue is virtually impossible at the moment.

As for the human element, you could argue that if we treated driving as the thing that it actually is - operating heavy dangerous machinery rather than a social, fun, emotional thing - and used the machines as instructed by the manufacturer, drive within the rules of the road and as instructed then there would (arguably) be far less crashes.

[Mr. Dazzle]

I know it's two different thing in the way it's written, but in terms of the fundamental goal/question here..."is it safe to let this person/machine out on the road?"...you're asking the same thing.

If you sat down and applied the same robustness and logic to the idea of letting a human drive, you'd probably conclude they shouldn't be allowed to It's just we all kind of accept it because "it's a person, innit?".

People are spectacularly crap at driving, nearly every accident which occurs is a result of driver error. it's very rare to have an 'unprovoked' car accident. If you tried to demonstrate people are reliable in the way you'd do it for a machine, you'd fail.

[iansoady]

What people are / can be good at is dealing with unpredicted situations. This is far more difficult to achieve with AI for fairly obvious reasons. Like the traffic cones on the bonnets of self-driving cars in California..... https://www.theguardian.com/us-news/202 ... reet-rebel

[Taff]

This thread isn't about the human in the loop though is it, it's about removing the human altogether and allowing self driving cars on the road.

All I've said, is that developing an acceptable safety argument for a self driving car is (IMO) damn near impossible at the moment because of the technology employed, and by extension developing good regulation for this new technology has problems around the sqep of the regulator. (Also consider anyone that has the sqep will be working for the manufacturer earning loads of money, not in the regulator on a civil service salary)

[Horse]

What people are / can be good at is dealing with unpredicted situations.

They often seem to struggle with 'normal' predictable events. If that wasn't the case then we wouldn't continue to see the same few types of gotcha bike crashes.

[Mr. Dazzle]

How often do unpredicted circumstances occur though, vs. entirely predictable crashes? Have you ever had a crash which someone else hasn't had before? Some people do have really weird scenarios but 99% of crashes are rinse-repeat of something someone has done before. Probably about 97% of them are the same 4 or 5 scenarios right?

I'd imagine you could dramatically reduce the overall number of accidents even if you accepted that 'weird' crashes would be more often. Of course that would mean accepting that more crashes of a certain kind will happen, even if the total rate dropped, which is an almost impossible decision.

EDIT: Further thought on this though - driver assistance aids are already focusing on this of course. Already we have cars which brake to mitigate rear end shunts, nudge you back over when you drift out of your lane, keep an eye on your blind-spots and pay attention to make sure you're not falling asleep or taking your hands off the wheel. Those must cover a huge percentage of all crashes.

[Horse]

How often do unpredicted circumstances occur though, vs. entirely predictable crashes? Have you ever had a crash which someone else hasn't had before?

I'd imagine you could dramatically reduce the overall number of accidents even if you accepted that 'weird' crashes would be more often.

This is an hour, so a fair commitment. If you don't have that spare, watch from 16:00, when it covers using historic crash data to develop, then vary, scenarios for testing AV software.

This means that a vast range of variations can be tested, that even a fleet of vehicles would be unlikely to encounter in on-road driving.

[The Spin Doctor]

I know it's two different thing in the way it's written, but in terms of the fundamental goal/question here..."is it safe to let this person/machine out on the road?"...you're asking the same thing.

If you sat down and applied the same robustness and logic to the idea of letting a human drive, you'd probably conclude they shouldn't be allowed to It's just we all kind of accept it because "it's a person, innit?".

People are spectacularly crap at driving, nearly every accident which occurs is a result of driver error. it's very rare to have an 'unprovoked' car accident. If you tried to demonstrate people are reliable in the way you'd do it for a machine, you'd fail.

Bring back trains.