The Self-Driving Vehicles Thread

[Horse]

Oh, and on driver monitoring, there may well eventually be systems like that in all cars, watching for tired drivers.

[Cousin Jack]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

How will it do that? Having hands on the wheel and looking ahead is no good, what is required is a brain monitoring system to ensure the driver is actually monitoring the situation in real time. And that is probably more difficult that creating a Level 5 car that doesn't need monitoring.

[wheelnut]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

So what’s the point and where’s the benefit?

[Horse]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

How will it do that? Having hands on the wheel and looking ahead is no good, what is required is a brain monitoring system to ensure the driver is actually monitoring the situation in real time. And that is probably more difficult that creating a Level 5 car that doesn't need monitoring.

Spin's challenge was 'what of the driver's asleep?' In which case, driver is less likely to be looking ahead and holding the wheel

Honest answer is that I don't know, I'm not involved with the work.

[Horse]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

So what’s the point and where’s the benefit?

For ordinary drivers on ordinary roads, I don't know. I'm not an evangelist for these systems, just answering the challenges where I can.

[iansoady]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

Now we've taken back control we won't have to follow those namby-pamby rules. Survival of the fittest (well richest....)

[The Spin Doctor]

See earlier in the thread. Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

How can it tell? Does it have some kind of state-of-mind metre to determine whether or not the driver will focus for the trip?

The driver could simply be gazing out a side window at the scenery.

As you know from the Startle Effect posts and the No Surprise! approach, handing control back to someone who's not expecting to have to 'fly the plane' often ends in disaster.

[Horse]

I think that, several times, I've said that the implementation most likely is first/last mile shuttle bus services.

https://techcrunch-com.cdn.ampproject.o ... twork/amp/

Ocado is treating this as a strategic investment to develop AI-powered, self-driving systems that will work across its operations, from vehicles within and around its packing warehouses through to the last-mile vehicles that deliver grocery orders to people’s homes. It says it expects the first products to come out of this deal — most likely in closed environments like warehouses rather the less structured prospect of open streets — to be online in two years.

[Horse]

if the driver isn't able to take back control.

How can it tell?

The driver could simply be gazing out a side window at the scenery.

As you know from the Startle Effect posts and the No Surprise! approach, handing control back to someone who's not expecting to have to 'fly the plane' often ends in disaster.

Having just added the previous post, I've noticed that my previous answer never made it to the board.

So take 2. But briefer than before.

How can it tell? Umpteen options, including head direction, eye direction, blink rate. Here's one system already installed by fleets.

https://www.exeros-technologies.com/sol ... on-system/



For startle, etc, that's the point of keeping the driver aware. The 'lipstick' example earlier would be a fail. Gazing (presumably you mean an extended look, more than looking at a door mirror) would possible fail too.

[The Spin Doctor]

Having just added the previous post, I've noticed that my previous answer never made it to the board.

So take 2. But briefer than before.

How can it tell? Umpteen options, including head direction, eye direction, blink rate. Here's one system already installed by fleets.

https://www.exeros-technologies.com/sol ... on-system/

I know all about that. There have been 'nudge' systems to refocus a driver for at least a decade. Falling asleep at the wheel - according to some research done about fifteen years back - causes three times as many crashes as alcohol, so there are systems that monitor blink rate and sound a klaxon to wake a driver up. And there are breath test lockout devices to stop someone who's had one too many from starting a journey in the first place.

But we're talking about a very different situation - as I am bloody certain you know very well indeed - and that's refocusing someone who isn't currently focused, to be able to take control.

You have to become alert, then analyse the situation, then decide what you're going to do about it. That takes seconds.

And as my recent posts on the Startle Effect show, we don't have seconds.

Either the vehicle is entirely reliable as an autonomous vehicle... or you let the driver drive with aids. The half-way house that says "nope, I'm not programmed for this, you have a go" is bonkers.

[Horse]

Falling asleep at the wheel - according to some research done about fifteen years back - causes three times as many crashes as alcohol, so there are systems that monitor blink rate and sound a klaxon to wake a driver up.

FWIW I think that may be part of future regs too, although not just blink rate.

But we're talking about a very different situation - as I am bloody certain you know very well indeed - and that's refocusing someone who isn't currently focused, to be able to take control.

You have to become alert, then analyse the situation, then decide what you're going to do about it. That takes seconds.

And as my recent posts on the Startle Effect show, we don't have seconds.

Either the vehicle is entirely reliable as an autonomous vehicle... or you let the driver drive with aids. The half-way house that says "nope, I'm not programmed for this, you have a go" is bonkers.

As I've said before, about other comments, the assumption seems to be that none of this has occurred to the developers, or that they're ignoring it. Do you really think that? [And, as I've also said before, UK not anywhere else in the world]

[The Spin Doctor]

As I've said before, about other comments, the assumption seems to be that none of this has occurred to the developers, or that they're ignoring it. Do you really think that? [And, as I've also said before, UK not anywhere else in the world]

Tesla.

[Horse]

As I've said before, about other comments, the assumption seems to be that none of this has occurred to the developers, or that they're ignoring it. Do you really think that? [And, as I've also said before, UK not anywhere else in the world]

Tesla.

Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment. While these features are designed to become more capable over time, the currently enabled features do not make the vehicle autonomous.

Full Self-Driving Capability
- Auto Lane Change: Assists in moving to an adjacent lane on the highway when indicated is engaged by driver
- Navigate on Autopilot (Beta): Actively augments Auto Lane Change by providing guidance to the driver to transit highway’s on-ramp to off-ramp, including suggesting lane changes and navigating interchanges
- Autopark: Helps parallel or perpendicular park your car, with a single touch
- Summon: Moves your car in and out of a tight space using the mobile app or key
- Smart Summon: Your car will navigate more complex environments and parking spaces, maneuvering around objects as necessary to come find you in a parking lot.


But it's nothing new for drivers to use vehicles beyond their capabilities.

Now, it could be fair to comment on their choice of names for the features, along with EM's statements, etc. Perhaps that's something that should be taken by anyone with concerns to the ASA.

However, in the UK there is:

https://www.gov.uk/government/publicati ... -trialling

Code of Practice: Automated vehicle trialling
Published 6 February 2019

Which includes:

Legal requirements

As stated in the introduction to this Code, conducting public trials of automated vehicle technology is possible in the UK at any level, provided the following legal requirements are met:

- a driver is present, in or out of the vehicle, who is ready, able, and willing to resume control of the vehicle

Vaguely similar to Tesla's statement?

The code goes on to:

Safety cases

Trialling organisations are expected to develop a detailed safety case before conducting trials which demonstrates that the trial activity can be conducted safely (with safety defined as the absence of unreasonable risk). Those conducting trials are expected to develop a safety case which is proportionate to the trial activity, and representative of any associated risks. The list below outlines the issues that should be covered by a safety case. It is not exhaustive, and those carrying out trials should build on this basic foundation:

information on the specific trial activity, vehicles, and operational domain of the trial

evidence that the trial activity can be performed safely, whether with a safety driver in the vehicle or with a remote safety operator

safety driver or operator training

processes for managing the trial activity, and organisational responsibilities for managing the trial

how the trial aligns with legislation and regulations

evidence of engagement with relevant bodies, authorities, and other road users

updates on milestones and progress reports of specific trial activity

Those carrying out trials are expected to develop and maintain a robust safety case and continue to assess the safety of any trial activity. Such safety cases are recommended to be made available to the public (see section 3.10).

Trialling organisations should consider safety standards and principles when developing a safety case, such as BSi PAS 11281:2018 (Connected automotive ecosystems. Impact on security of safety – code of practice).

Which may be where Tesla have their own view? Is there a published safety case? I don't know, I'm not involved much with this aspect. But that does emphasise what I have said about the UK's expectations.

Google finds this:

https://www.tesla.com/en_GB/VehicleSafetyReport

And

https://www.thatcham.org/tesla-autopilot-ruling/

[The Spin Doctor]

Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment.

Circular argument.

Clearly, the cars ARE out there on the road MINUS the systems that protect the driver AND other road users.

[Horse]

Forthcoming vehicle regs (if the uk decide to parallel the EU) will monitor the driver and won't allow the vehicle to self-drive if the driver isn't able to take back control.

Autopilot and Full Self-Driving Capability are intended for use with a fully attentive driver, who has their hands on the wheel and is prepared to take over at any moment.

Circular argument.

Clearly, the cars ARE out there on the road MINUS the systems that protect the driver AND other road users.

IIRC Teslas have a detection system, although drivers are obviously able to circumvent it.

But it's nothing new for drivers to use vehicles beyond their capabilities.

To clarify, I meant what the vehicle is designed to do. That said, we have plenty of instances (riders in bends?) where the vehicle operators run out of talent.

[Cousin Jack]

So Tesla have a non-foolproof fail-safe system. Or to put it bluntly, it dosen't work.

And as for 'riders in bends', yes humans are very fallible. So why give them an extra system to fail with, the existing car/bike is flawed but we have used the last 100 years to develop it to be much safer than it was. Perhaps the effort should go into continuing that process, rather than inventing a brand new way for cars to kill humans.

[Horse]

we have used the last 100 years to develop it to be much safer than it was.

Even that apparent improvement has been challenged. Eg risk compensation from safer vehicles having adverse effects on vulnerable road users.

Perhaps the effort should go into continuing that process, rather than inventing a brand new way for cars to kill humans.

Yet again, there seems to be a presumption that isn't happening too. It is.

[The Spin Doctor]

IIRC Teslas have a detection system, although drivers are obviously able to circumvent it.

You are still missing - or ignoring - the point.

If an emergency develops AND the vehicle needs to hand back control to a driver, to bring that driver up to the level of cognitive awareness needed to deal with the emergency takes SECONDS.

And that's far too long on the road.

However you parse it, a system that allows a driver to be hands-off and relies on the driver to remain alert is deeply flawed.

[Horse]

IIRC Teslas have a detection system, although drivers are obviously able to circumvent it.

You are still missing - or ignoring - the point.

If an emergency develops AND the vehicle needs to hand back control to a driver, to bring that driver up to the level of cognitive awareness needed to deal with the emergency takes SECONDS.

And that's far too long on the road.

However you parse it, a system that allows a driver to be hands-off and relies on the driver to remain alert is deeply flawed.

Ah, now you're veering into 'trolley bus' territory, with the assumption that the vehicle will commit itself to those 'surprise' situations. Oddly enough, there's a lot of work going on around preventing that. But, obviously, it will happen. If the benchmark was perfect drivers who never ever get caught out, then fair enough. As CJ pointed out, humans are fallible.

If - big 'if' - the system is usually better than a human and can alert the human in sufficient time when required, how would you feel about that?

Check your email.

[Cousin Jack]

If - big 'if' - the system is usually better than a human and can alert the human in sufficient time when required, how would you feel about that?

Not a big 'if', a f---ing HUGE 'if', and one most unlikely to be fixed ever.

If it could recognise the problem that far ahead it could just stop, so almost by definition the 'problem' scenarios where it can't deal with stuff will be rather urgent. The reality is that cars will hand back control when a crash s inevitable, so just shifting the blame really.