Tesla and Uber Unlikely to Survive (Vol. 2)

[gangzoom]

LimJim said:

Very impressive, but what is test driver playing at. No excuse for not having both hands on the wheel with a cyclist on either side.

I think he is holding the wheel at the 6oclock position, some of the FSD beta bloggers 100% don't hold the wheel though.

The way the car saw and than treated the kid cycling towards the car as any human drivers would do was very impressive. People here talk about kids jumping out in front of these cars and causing them to crash, that's clearly not going to happen. There will be a crash, but in the same way if a person decides to step into front of a moving car right now.

There is another video of the car stopping for an old lady crossing the road slowly, as you would do if you were driving. So you can make the car stop if you stop in the middle of the road, but you can also do that right now.....I doubt any one here will mow down a pedestrian standing in the middle of the road on their commute to work today?

The decision making seems to be improving very quickly now, it can only be a matter of time before this software is pushed out fleet wide. Am 100% there will be accidents and sadly some of them fatal, but every time any one drives a car there is a chance you will be involved in a fatal accident, what matters is the size of the denominator.

[98elise]

jsf said:

ZesPak said:

jsf said:

These electric modern cars are full of expensive electronics, cost a fortune to repair post crash and i bet the electronics and wiring looms external to the cabin hate salted winter roads.

efa

Indeed, which doesn't negate the point I was making, that the poster I was replying to was typing nonsense.
These Electric cars wont last any longer than an ICE.

I disagree. I've spent decades maintaining and fixing all manner of complex electronic, electric and mechanical systems (from missiles to air conditioning).

Electrical and electronic systems beat anything mechanical hands down for reliability.

[Tuna]

gangzoom said:

I think he is holding the wheel at the 6oclock position, some of the FSD beta bloggers 100% don't hold the wheel though.

The way the car saw and than treated the kid cycling towards the car as any human drivers would do was very impressive. People here talk about kids jumping out in front of these cars and causing them to crash, that's clearly not going to happen. There will be a crash, but in the same way if a person decides to step into front of a moving car right now.

There is another video of the car stopping for an old lady crossing the road slowly, as you would do if you were driving. So you can make the car stop if you stop in the middle of the road, but you can also do that right now.....I doubt any one here will mow down a pedestrian standing in the middle of the road on their commute to work today?

The decision making seems to be improving very quickly now, it can only be a matter of time before this software is pushed out fleet wide. Am 100% there will be accidents and sadly some of them fatal, but every time any one drives a car there is a chance you will be involved in a fatal accident, what matters is the size of the denominator.

I think what you're seeing is better object detection - which is what the full re-write was meant to be about - not better decision making.

When we talk about decision making, it's not "shall I mow down the old lady or brake to a halt?", it's "how do I interpret this ambiguous scene where it's not clear what I and other drivers should do?". Steering smoothly around an object or stopping for it, whether it's a child on a bike or a granny crossing the road is not new. We were solving dynamic path finding in the nineties. No AI involved at all.

[gangzoom]

Tuna said:

We were solving dynamic path finding in the nineties. No AI involved at all.

Can you show us some of your work from the 90s?

Loads of people on here claim they can do all sort of stuff, most of the time its rubbish, but occasionally its ture. Am more interested in how any of this tech is developing rather than the end product. From my own personal work point of view we are now getting to the point where we have so much data on a person interms of medical treatment, presenting that data to a human is becoming almost impossible. The next step is how we create actual semi intelligent work process that can automate data processing so we can free up human brings to do the more complex stuff.

Sadly the world of clinical medicine is so far behind the automative world on this front its not even funny, even Imperial who have a direct partnership with DeepMind aren't light years ahead. If there is software that can guide a car through a series of risk based assessments/judgements there is no reason why we cannot use the same techniques to help reduce human errors in clinical medicine - which occurs on a daily basis, none of it is intentional.

Edited by gangzoom on Thursday 17th December 08:26

[Castrol for a knave]

jsf said:

ZesPak said:

jsf said:

These electric modern cars are full of expensive electronics, cost a fortune to repair post crash and i bet the electronics and wiring looms external to the cabin hate salted winter roads.

efa

Indeed, which doesn't negate the point I was making, that the poster I was replying to was typing nonsense.
These Electric cars wont last any longer than an ICE.

I offered a view, not saying it's going to be 100%, just a view.

I even finished my ramblings with a comment that I have been known to get things wrong. I wouldn't say it was nonsense, none of us know where we will end up in 10 or 15 years. We could all be wrong, even you.

[jamoor]

98elise said:

jsf said:

ZesPak said:

jsf said:

These electric modern cars are full of expensive electronics, cost a fortune to repair post crash and i bet the electronics and wiring looms external to the cabin hate salted winter roads.

efa

Indeed, which doesn't negate the point I was making, that the poster I was replying to was typing nonsense.
These Electric cars wont last any longer than an ICE.

I disagree. I've spent decades maintaining and fixing all manner of complex electronic, electric and mechanical systems (from missiles to air conditioning).

Electrical and electronic systems beat anything mechanical hands down for reliability.

Indeed, this electronics myth has been perpetuated by those that couldn't get their head around EFI when it came around and have been saying "modern cars too many electronics" since.

[Justin Case]

Most modern cars are scrapped because a: a lot of components are non-repairable and expensive to replace, and b: because of the way cars are assembled, no thought is given to dis-assembly for repairs/replacements, making the task complex and costly. Unfortunately I don't see this becoming any cheaper with EVs, so even if they could last longer, they won't

[jamoor]

Justin Case said:

Most modern cars are scrapped because a: a lot of components are non-repairable and expensive to replace, and b: because of the way cars are assembled, no thought is given to dis-assembly for repairs/replacements, making the task complex and costly. Unfortunately I don't see this becoming any cheaper with EVs, so even if they could last longer, they won't

This depends entirely on which car the country is located.


If you can get a mechanic in India for £5-10 a day he will spend all day fixing a car worth £1000

Edited by jamoor on Thursday 17th December 10:13

[Richard-D]

Justin Case said:

Most modern cars are scrapped because a: a lot of components are non-repairable and expensive to replace, and b: because of the way cars are assembled, no thought is given to dis-assembly for repairs/replacements, making the task complex and costly. Unfortunately I don't see this becoming any cheaper with EVs, so even if they could last longer, they won't

No no no! Listen to Jamoor, cars are only ever scrapped because the ICE components fail. The rest of the vehicle will be absolutely fine.

[jamoor]

Richard-D said:

Justin Case said:

Most modern cars are scrapped because a: a lot of components are non-repairable and expensive to replace, and b: because of the way cars are assembled, no thought is given to dis-assembly for repairs/replacements, making the task complex and costly. Unfortunately I don't see this becoming any cheaper with EVs, so even if they could last longer, they won't

No no no! Listen to Jamoor, cars are only ever scrapped because the ICE components fail. The rest of the vehicle will be absolutely fine.

Only when it’s worth £500 do you scrap it.
You wouldn’t scrap a car that’s worth 10k would you?

[Tuna]

gangzoom said:

Can you show us some of your work from the 90s?

"We" as in the software industry. Pixar were doing some very nice things with flocking, path finding and dynamic routing - there are a whole bunch of papers they wrote on it. The games industry were all over it, as the transition into 3D meant than NPCs had to do more than just bounce around. Dynamically choosing routes that were robust, accurate, repeatable and 'naturalistic' was a major area of research, not just in the entertainment industry, but in simulation and military operations.

At the time, none of it was put into cars because the sensor technology wasn't there - that's the "secret sauce" that has unlocked the various attempts at self-driving vehicles: the ability to process a scene and decompose it into targets to react to in real time.

[Tuna]

gangzoom said:

Sadly the world of clinical medicine is so far behind the automative world on this front its not even funny, even Imperial who have a direct partnership with DeepMind aren't light years ahead. If there is software that can guide a car through a series of risk based assessments/judgements there is no reason why we cannot use the same techniques to help reduce human errors in clinical medicine - which occurs on a daily basis, none of it is intentional.

I've worked in exactly this space - and the problem is the technology is not generalisable, and not robust. You can't take a "risk assessor" from a car and put it into a heart monitor. Each piece of software is highly specialised and tuned to a very specific environment. And when you take the general techniques we're using at the moment (neural nets, deep learning and so on), you find out that the results can be statistically acceptable, but dangerously unpredictable in a high risk environment. It's like having a doctor that gives the right diagnosis ninety nine times out of a hundred, but that other one time will prescribe arsenic.

[LimJim]

What's going on with guidance YTD? Havn't been following but just past a tesla garage and seemed busier than I've every seen it.

[hyphen]

LimJim said:

What's going on with guidance YTD? Havn't been following but just past a tesla garage and seemed busier than I've every seen it.

Doesn't mean they will sell any

"Hello, I just went past, and want to impulse buy a Tesla for my partner for christmas"
"We don't have any to buy in stock, We only ship them over every three months from the USA"
"Oh never mind, will get something else"

[LimJim]

I don't think it's that kind of garage. More like a parking lot where you pick up internet orders.

[hyphen]

jamoor said:

This depends entirely on which car the country is located.
If you can get a mechanic in India for £5-10 a day he will spend all day fixing a car worth £1000

That mechanic from India can be found in the back streets garages of the UK

[Order66]

LimJim said:

What's going on with guidance YTD? Havn't been following but just past a tesla garage and seemed busier than I've every seen it.

December is delivery month - literally thousands being delivered this month. Tesla have rented massive car parks across the country to manage the deliveries.

As for waiting 3 months for delivery - would be a very unusual case. There's people who ordered last week picking up this week. They have a certain amount of "inventory", you'd only need to wait the 3 months if your timing was bad and you had unusual spec needs.

[ZesPak]

Order66 said:

As for waiting 3 months for delivery - would be a very unusual case. There's people who ordered last week picking up this week. They have a certain amount of "inventory", you'd only need to wait the 3 months if your timing was bad and you had unusual spec needs.

As far as I'm aware they still don't produce anything by order.
If my math is correct, they got about 40 configurations in total to deliver, that is taking into account Model, exterior colour, tow hitch and interior colour.

[jamoor]

hyphen said:

That mechanic from India can be found in the back streets garages of the UK

for double the price

[RichardAP]

Order66 said:

LimJim said:

What's going on with guidance YTD? Havn't been following but just past a tesla garage and seemed busier than I've every seen it.

December is delivery month - literally thousands being delivered this month. Tesla have rented massive car parks across the country to manage the deliveries.

As for waiting 3 months for delivery - would be a very unusual case. There's people who ordered last week picking up this week. They have a certain amount of "inventory", you'd only need to wait the 3 months if your timing was bad and you had unusual spec needs.

I’m interested in how they cope with doing PDI’s on a feast to famine basis, must be hard to keep an experienced team around with nothing to do for 2 months then rushed off or their feet, anyone know?