Tesla autopilot fatal crash

[tazzieman]

It's almost like this technology is improving ??

https://electrek.co/2016/12/27/tesla-autopilot-radar-technology-predict-accident-dashcam/

Idiot in the hatchback could've killed someone. Goes to show just cos you drive a big 4wd you aren't safe against some little ankle biting slalom pilot. 

[JWM]

Indeed, video was nuts! Lucky no one was killed. 

[tazzieman]

https://www.reuters.com/article/us-tesla-crash-idUSKBN19A2XC

The NTSB report disclosed that the Tesla Model S uses a proprietary system to record a vehicle's speed and other data, which authorities cannot access with the commercial tools used to access information from event data recorders in most other cars.

For that reason, the NTSB said it "had to rely on Tesla to provide the data in engineering units using proprietary manufacturer software."

I foresee black box technology in all cars of the future. Your private driving habits will be viewable by anyone with the authority to do so. Thus retrospective speeding fines might be on the cards. Don't think it couldn't happen. Anyone using a smartphone is already being monitored if not actively then potentially retrospectively. 1984 is here!

[Redracn]

Looks like hacking an autonomous vehicle will be far easier than imagined. What could be done with signs on the back of your own car I wonder. 

https://www.wheelsmag.com.au/news/1708/autonomous-cars-can-be-hacked-with-stickers-report

[Coastr]

You don't get endless funds from the government without giving up information in return.  Remember when apple made a big show of not unlocking a terrorists phone?  Yeah no the CIA et al have lots of backdoors as revelead by whistleblowers.  Just a bit of theatre for the masses.  Same here as tesla.

no big deal though insurance companies can already get the black box in your Holden to see if you were WOT on the back straight or really swerving to avoid a kangaroo. 

[Redracn]

When a not very autonomous car crashes it is apparently still human error.

https://www.wheelsmag.com.au/news/1709/tesla-shoulders-some-of-the-blame-for-fatal-autopilot-crash

Tesla should accept full responsibility for the failure of its car and we need laws like we have in other areas of engineering to make them responsible. That way they may deliver a product that actually works. 

[LeeM]

'These systems require the attention of the driver at all times'

 Incase the failsafe autonomy fails, and you wonder why you bought an autonomous car in the first place?  Sure, have collision warnings so that the numpty drivers who are oblivious to other road users are aware they are about to have an accident, and can react accordingly (probably not coz they're stupid), yet to rely on a computer just makes you an idiot

[Redracn]

The one thing you will see on almost all electronic compent data sheets is a warning stating that the product is "Not for use in Life Support". Before I place my life in the hands of a complex machine I would want to know they every sensor/computer/software/connector etc was rated as suitable for use in Life Support. 

[tazzieman]

 Before I place my life in the hands of a complex machine I would want to know they every sensor/computer/software/connector etc was rated as suitable for use in Life Support. 

So you trust taxi drivers?  I don't!

[Redracn]

So you trust taxi drivers?  I don't!

I avoid those death traps (Car and Driver) whenever I can. I try to use Limos/Private hire cars which are not much more expensive and way better drivers/cars/service.

[tazzieman]

I avoid those death traps (Car and Driver) whenever I can. I try to use Limos/Private hire cars which are not much more expensive and way better drivers/cars/service.

What if I picked you up from the airport in my

I have a 38 year clean driving record...but have encountered countless nongs along the way

[Redracn]

What if I picked you up from the airport in my

I have a 38 year clean driving record...but have encountered countless nongs along the way

Would have to pass as no headrests otherwise looks good? 

[tazzieman]

Would have to pass as no headrests otherwise looks good? 

Ah but that is not mine. Mine has Saab headrests that I installed. The 60s Mercs however don't yet , but I am driving them around without fear.

[Oldmxnut]

Hi Gents, reading through the comments todate and holding off commenting as sometimes you just have to be quite but its a Monday evening and Im in a sharing mood .. Apologies. 

Im in automous mining (open pit), in partnership with an American company to put following comments into context.  

I own and operate a number of autonomous mining trucks of which presently are driver assist which means the driver also is in the truck (although not really required).  (No we are not Rio Tinto, although you would think by their marketing they were the original inventors)   The technology is here and has been for some time remembering we have had auto pilot commercial planes for ages for example.

One (of a few) key challenge for this technology is the bridging the "commercially realistic and socially acceptable gap".   The cost of this technology that would have prevented this recent event is say $500K US for example.  This is close to being an acceptable marginal cost on a $3Million mining truck but isn't possible on a $50K car.  So what ends up happening is the car companies including tesla are trying to derive a cheaper more accessible form of which pieces are missing. Our US partners work with a major car OEM and  its a big issue.   Think the size and cost of computers 20 years ago and the size and cost of an iPhone today.  Computers then could possibly do everything an iPhone could do but they were the size of a room and cost a million.   Someone will crack this code.. who knows maybe tesla ?  Maybe a steve Jobs version 2 somewhere in the world... and then in a blink of any eye nearly every mode of people transport will change.  

Second challenge is the philosophical question about what we will accept as collateral or social cost.   Imagine getting on a plane to LA tomorrow morning knowing that there is no pilots on the plane at any stage.  It is possible right now but we wouldn't accept that as a society, although is majority of events, autonomous pilot would deal better with circumstance then human pilot.   Also the other question i like to ask.. "who dies and who decides".   We drive down the road tomorrow and weave to miss the middle aged over weight guy waiting for a heart attack stepping onto the road and hit head on into the young child bearing mother in the on coming car or skid and hit the guy stepping out on the road.   What ever happens as long as we didn't do something illegal, the judge puts it down to "human error" and an accident.   With mathematical circumstances, there is never an accident, an autonomous vehicle would have determined who the individual was as he stepped out on the road and already knows the driver of the oncoming vehicle and a mathematical formulas determined if the physical could not be over come (avoiding any accident) then decides if the cars hits the man or the young woman.   The question is then, who provides that formula and secondly do you then rely on AI (artificial intelligence) to determine that..   hence the big big question for humanity...

And just to confirm the technology and apparatus is available right now to do exactly what i describe in the above philosophical question.  

To comment further on the Tesla incident.  The apparatus and technology is available and possible for the vehicle to have determined that the truck wasn't giving way as he was supposed to do and taken avoidance action and most probably much quicker and more effectively then a human could have.  The problem I believe is that at this time, this technology and hardware is not practical (cost and size) to put on a $50K tesla, per my earlier comments.   Hence my social acceptable comment, at what price and risk will we accept the technology in our vehicles and what would we pay for it, but ultimately at some point I believe this gap will be bridged. 

 

 

Edited 18September, 2017 by Oldmxnut

[Redracn]

^^^

I think you will find that the highly regulated mining pit which a small area with all the required  expensive infrastructure to support your autonomous equipment is not in any way representative of the general automotive world. To be viable at L5 an autonomous car will have to identify every possible object around it both mobile and static and get it right every time so a decision can be made. Brick fence or colourbond etc. I bet your expensive state of the art trucks can not do this. But as the car owner I would expect it to protect the occupants (my family) first otherwise why bother with it. I won't be touching one until they get Kangaroos right and can detect still water over a road at night. They will probably get there in the end but not anytime soon and if it works as well as the much simpler task of voice recognition then god help us all because voice recognition still sucks after 30+ years of development. 

[bumble]

An article in the weekend fish and chip wrappers (can't remember which one) discussed this subject.  The takeaway was that even after decades of development in aviation, auto-pilots still have problems.  Planes still require experienced biological units to be present.  Mining trucks were mentioned and even in a highly controlled more or less static environment these things are still monitored closely by humans in real time at Rio T.  Quite clearly, there are so many social and technical issues to overcome, everyone needs to take a Bex and have a good lie down regarding fully autonomous vehicles operating in the chaos of a city environment because it's unlikely to happen in any of our lifetimes.  Meanwhile, the likes of Apple, Google, Amazon, Tesla, Uncle Tom Cobley and all will dazzle the gullible and ignorant with demos of their latest pod thingy whirring around some retirement village at 3km/hr getting in the way of the Zimmer frame set.  And the millennials will wake up in middle age and wonder why their Uber still has a driver...

A pox on them all - time for my cocoa...

[smit2100]

An article in the weekend fish and chip wrappers (can't remember which one) discussed this subject.  The takeaway was that even after decades of development in aviation, auto-pilots still have problems.  Planes still require experienced biological units to be present.  Mining trucks were mentioned and even in a highly controlled more or less static environment these things are still monitored closely by humans in real time at Rio T.  Quite clearly, there are so many social and technical issues to overcome, everyone needs to take a Bex and have a good lie down regarding fully autonomous vehicles operating in the chaos of a city environment because it's unlikely to happen in any of our lifetimes.  Meanwhile, the likes of Apple, Google, Amazon, Tesla, Uncle Tom Cobley and all will dazzle the gullible and ignorant with demos of their latest pod thingy whirring around some retirement village at 3km/hr getting in the way of the Zimmer frame set.  And the millennials will wake up in middle age and wonder why their Uber still has a driver...

A pox on them all - time for my cocoa...

I watched a rerun of a movie on the weekend where a very greying pilot ditched a jumbo in the  River in New York back a decade or so ago shortly after takeoff and the engines getting wiped out by a flock of stray birds.  Not sure how real the aviation inquiry was they referenced in the movie  compared to reality, where they ran simulations after the fact to try and string the pilot and co-pilot for unnessarily ditching the plane, but lots of assumptions made that were not reality in the post simulations.  eg change a parameter / assumption here or there in real time and the outcome on the autopilot is totally different.

That said, I still  see cars of the future not too far off being void of a pedal that most cars of today take for granted.  Eg Only have one pedal in the future, a brake pedal.  It will  be a autopilot override,  in case those bot's, AI machine learnings and programmers make the odd error in those algo's in terms of getting their 1's and 0's right.  Also reckon the outlaw of the future will be mainly those running outlawed autopilot 1's and 0 tunes that overrides the sensors, gps trackers, cameras and prying eyes telemetry and has the single pedal acting as hybrid with a bit of back to the future, i.e acts as both a brake pedal and accelerator.

PS On the topic of static environments, I was in traffic and late to a meeting with a van full of colleagues on the streets of Mumbai and the topic of autonomous cars came up.  I just looked out the window and sighted varying road level grading, awesome lane discipline and lane markings defining the boundaries of what constitutes the actual road and having the need for an awesome eye / depth perception in advance in the third  dimension (evergreen and ever growing potholes) and thinking on top of that, how the heck do you program for anticipating what constitutes a hazard / risk averse behavior of others  (what are those 5 tuk tuk drivers thinking and how about those three recalcitrant pedestrians with no concept of looking  left , then right then  left, eg try getting them to just look before stepping out in front / taking on a 2.0 + tonne lightweight.   We all thought, not in our lifetime,  for Mumbai at least. 

Edited 18September, 2017 by smit2100

[Redracn]

An article in the weekend fish and chip wrappers (can't remember which one) discussed this subject.  The takeaway was that even after decades of development in aviation, auto-pilots still have problems.  Planes still require experienced biological units to be present.  Mining trucks were mentioned and even in a highly controlled more or less static environment these things are still monitored closely by humans in real time at Rio T.  Quite clearly, there are so many social and technical issues to overcome, everyone needs to take a Bex and have a good lie down regarding fully autonomous vehicles operating in the chaos of a city environment because it's unlikely to happen in any of our lifetimes.  Meanwhile, the likes of Apple, Google, Amazon, Tesla, Uncle Tom Cobley and all will dazzle the gullible and ignorant with demos of their latest pod thingy whirring around some retirement village at 3km/hr getting in the way of the Zimmer frame set.  And the millennials will wake up in middle age and wonder why their Uber still has a driver...

A pox on them all - time for my cocoa...

^^ Sums it up well. 

There should be a checklist of functions the autonomous car needs to perform before being given the honour of being called autonomous.

We could start with overtaking. Particularly on winding hilly country roads with poor/no line markings. Would the speed limit be non negotiable or could it be exceed for safety reasons. (This one is probably way to hard so no overtaking will be allowed)

Then move to roadworks. Anything from a simple tree lop/powerline job to the need to travel on both sides of the road with no line markings on varying surfaces and handle detours around construction. It needs to navigate this correctly with no cars to follow. Would need to obey both lights and the traffic controller (or any worker) who if you are the first car in line could be right next to your window and give you the nod to go or rotate the sign which is right next to and above your roof.

Next test could be an accident scene. Would need to spot the accident which may not even be on the road but off to the side. Will it then stop and render assistance? Secue the scene?Perform CPR? Wave down other cars? Etc etc.

Animial detection. Can it identify an animal from any angle including laying down using only a part of its body due to obstructions obscuring it. Can it then work out the risk based on the animal, time of day and time of year. Can it correctly identify any barrier between you and the animal and evaluate its effectiveness at keeping the particular animal out of its path? Multiple different and similar animals both sides of the road all at once. How many sheep (people) can it track at once in all their varying forms?

Follow the directions of a police officer. But of course it needs to identify that person as a police officer first.

Read and understand all road advisory signs including temporary ones and place appropriate limits on its calculations in accordance with the advice. 

It will need to be hack proof. This is even more important if it relies on communicating with other cars and infrastructure as it would be easy to supply incorrect information such as lights changing to red in 10seconds but it happens now or all lights green when they are red or a parked car saying it has crashed in the middle of the intersection.  

etc

etc

The list will be very very long. So as stated previously you will not see a true autonomous vehicle anytime soon. 

 

[KGB]

I observe a plethora of dickheads every morning and evening engaged in autonomous iPhone drive.  Apparently no need for constant eyes on the road with that 'system'.

[Oldmxnut]

^^^

I think you will find that the highly regulated mining pit which a small area with all the required  expensive infrastructure to support your autonomous equipment is not in any way representative of the general automotive world. To be viable at L5 an autonomous car will have to identify every possible object around it both mobile and static and get it right every time so a decision can be made. Brick fence or colourbond etc. I bet your expensive state of the art trucks can not do this. But as the car owner I would expect it to protect the occupants (my family) first otherwise why bother with it. I won't be touching one until they get Kangaroos right and can detect still water over a road at night. They will probably get there in the end but not anytime soon and if it works as well as the much simpler task of voice recognition then god help us all because voice recognition still sucks after 30+ years of development. 

Don't get me wrong, I'm not a big advocate for autonomous public vehicles, but I think maybe the point I was trying to make was missed.  The point is that there is technology (software and hardware) right now that does all of that and more.  Even the notes around road works and adverse outcome determinations.  But the missing pieces between it being on a car today or at some point in 30 years time... is the commercial cost and practicality, along with social acceptance of a different risk profile, not the technology.  Just don't make the assumption that the technology is still to be invented.  When someone cracks the cost / benefit problem and it will turn up very quickly, subject to the final hurdle of social acceptance.

To pick up on a few of your points:

Determining a brick fence and color bond is relatively easy.   The technology we have on the trucks does this and yes picks the difference between kangaroos and people, standing, lying down, running, walking whatever, and this is still very basic in the scheme of what is possible.  There are further options where it maps the walls and roads, the rocks types and yes also the moisture, surface or otherwise along with temperature etc.. the list goes on.  We don't use because again of the cost/benefit and practicality.  These are just all the simple things.    As I said, the more complex and what we really should be getting concerned about in my opinion is the ''thinking" part and what decisions are made with the information.  

The other point around social acceptance.  Your point around safety and family is spot on.  So lets assume there may be a day where all of these things can be put on a vehicle commercially and effectively but still we (including myself) would still be hesitant to "let go".   But we still would much rather accept the risk of driving down the road at 100 km passing oncoming cars with a meter of tolerance, not knowing the condition of vehicle and the state of the driver, yet wouldn't accept a machine with all of the above technologies doing the same.  That maths whizzes would prove up that how much lower the probability of something going wrong would be, but I agree, we would still be hesitant.    I personally think until we progressively pass 1 million autonomous vehicles coming the other way will we accept it, just as for the last 20-30 years we personally have safely passed a million vehicles without incident thereby conditioning ourselves that its safe to do so.    I still have a funny feeling when standing beside a road with a driver-less truck coming down it... trust me its a weird feeling irrespective of what your brain is telling you.

Last point, voice recognition is a good example of the above conundrum.  None of us want to pay any extra for VR on our vehicles, therefore the one in our car is the cheap import so to speak sadly.  Its definitely not the one that the FBI have to pick out your voice out of millions of mobile calls in a second and have it stand up up in a court of law that it belongs to you and nobody else in the world... cost / benefit and practicality.

Hopefully nobody takes offence at my comments as Im with the majority and cautious crowd, not so much as I don't trust the current technology but where its going with AI.  

 

 

[Oldmxnut]

^^ Sums it up well. 

There should be a checklist of functions the autonomous car needs to perform before being given the honour of being called autonomous.

We could start with overtaking. Particularly on winding hilly country roads with poor/no line markings. Would the speed limit be non negotiable or could it be exceed for safety reasons. (This one is probably way to hard so no overtaking will be allowed)

Then move to roadworks. Anything from a simple tree lop/powerline job to the need to travel on both sides of the road with no line markings on varying surfaces and handle detours around construction. It needs to navigate this correctly with no cars to follow. Would need to obey both lights and the traffic controller (or any worker) who if you are the first car in line could be right next to your window and give you the nod to go or rotate the sign which is right next to and above your roof.

Next test could be an accident scene. Would need to spot the accident which may not even be on the road but off to the side. Will it then stop and render assistance? Secue the scene?Perform CPR? Wave down other cars? Etc etc.

Animial detection. Can it identify an animal from any angle including laying down using only a part of its body due to obstructions obscuring it. Can it then work out the risk based on the animal, time of day and time of year. Can it correctly identify any barrier between you and the animal and evaluate its effectiveness at keeping the particular animal out of its path? Multiple different and similar animals both sides of the road all at once. How many sheep (people) can it track at once in all their varying forms?

Follow the directions of a police officer. But of course it needs to identify that person as a police officer first.

Read and understand all road advisory signs including temporary ones and place appropriate limits on its calculations in accordance with the advice. 

It will need to be hack proof. This is even more important if it relies on communicating with other cars and infrastructure as it would be easy to supply incorrect information such as lights changing to red in 10seconds but it happens now or all lights green when they are red or a parked car saying it has crashed in the middle of the intersection.  

etc

etc

The list will be very very long. So as stated previously you will not see a true autonomous vehicle anytime soon. 

 

To answer some of your questions:

- Autonomous is being used as a marketing tool for companies with some aspects therefore it wrongly described in a number of cases.

- Overtaking, speeds, roadworks,animals, humans, trees, bikes, scooters, any other moving object, random events, things falling out the sky, birds, rain event as compared to someone spraying water on road, tyre wear, braking distance, tolerance in mechanical moving pieces, crashes and then what responses other vehicles or emergency paths required, occupants changing there minds, toilet breaks, heart attacks in vehicle, motion sickness the list is infinite. 

But all determinable... colours, density, size,shape, distance, speed, projected paths, trajectory, response analytics being computed at millions of times faster then our brain can via a range of different technologies all available right now. 

We will see progressively more and more  vehicles with more and more aspects of "Autonomy" . This also overlaid with progressively less controlled environments.  Have a think about what is on current vehicles right now and at relatively fairly cheap entry points.  Then consider what we had on the same vehicles, 5 and 10 years ago and noting that like everything before, the development curve is exponential usually at some point and not straight line.

I personally think we will see a majority of personal vehicles with a driver less option electable (or probably compulsory) for certain environments within the next 10 years.  Eg, Picture our car, where we drive and entry a major free way and at the on ramp, it goes autonomous. 

 

 

 

 

 

 

 

 

[Redracn]

I have to disagree that the technology is ready and it is about cost. There is a huge way to go with the technology. I know a bit about electronics and if the tech is ready then cost reduction is a no brainer and not difficult given the volumes involved. But they need to know what to do so they can design the silicon to do the specific task. If the tech is ready as you say then why aren't they testing L5 vehicles on public roads right now after all cost is no object to the development teams. 

VR still gets it wrong even today more than 30 years after they started on it. I also mean all VR not just cheap units even the alleged great FBI stuff (recording something for voice printing is not VR and grabing a few keywords is a limited form of VR). Perhaps to many scifi TV shows. Getting VR wrong has little to no consequence except annoying the user. Getting it wrong in an Autonomous vehicle could kill you. 

If the tech is done as you say then why don't we have fully autonomous robotic home assistants as cost is not an object for the early adopters in this area. 

The tech is not ready at any price. I bet your trucks have a long list of operating restrictions and conditions.? When your truck writes its own operators manual without human input I will be impressed. 

[tazzieman]

The boffins will not be able to work around kangaroos and the likes of 

I'm going to have a lot of fun trolling the next generation.

[Redracn]

To answer some of your questions:

- Autonomous is being used as a marketing tool for companies with some aspects therefore it wrongly described in a number of cases.

- Overtaking, speeds, roadworks,animals, humans, trees, bikes, scooters, any other moving object, random events, things falling out the sky, birds, rain event as compared to someone spraying water on road, tyre wear, braking distance, tolerance in mechanical moving pieces, crashes and then what responses other vehicles or emergency paths required, occupants changing there minds, toilet breaks, heart attacks in vehicle, motion sickness the list is infinite. 

But all determinable... colours, density, size,shape, distance, speed, projected paths, trajectory, response analytics being computed at millions of times faster then our brain can via a range of different technologies all available right now. 

We will see progressively more and more  vehicles with more and more aspects of "Autonomy" . This also overlaid with progressively less controlled environments.  Have a think about what is on current vehicles right now and at relatively fairly cheap entry points.  Then consider what we had on the same vehicles, 5 and 10 years ago and noting that like everything before, the development curve is exponential usually at some point and not straight line.

I personally think we will see a majority of personal vehicles with a driver less option electable (or probably compulsory) for certain environments within the next 10 years.  Eg, Picture our car, where we drive and entry a major free way and at the on ramp, it goes autonomous. 

 

 

As you say the list of situations is infinite and as such is NOT determinable as you also state. Computers are eaisly programmed to handle specific situations but it takes true AI (not the rubbish they are calling AI) to handle the infine and unknow by the programmer situations. So until they have true AI and visual processing capability that exceeds that of humans a true Autonomous car is just a pipe dream. As you say they currently have some aspects of automation which are best described as driver assistance systems. Calling them autonomous is just wrong. More capability will as you say be rolled out over time as they head to the final target but there are many issues that could stall progress for decades or longer, technical, regulatory and legal

I have had a number of cars with collision avoidance and they have all had major issues. One of the worst was comming to an emergency stop in the middle of an intersection with my foot still on the throttle. Fortunately no other cars were anywhere near me as anybody behind whould have been toast. This tech while it may assist on occasion is not really ready for prime time. 

As far as I see it drivers with very good risk perception skills will be at a disadvantage with autonomous vehicles as they could do better while the dumbest least skilled drivers will benefit. This just seems fundamentally wrong in an evolutionary sense. Our survival instinct will become a thing of the past. 

The boffins will not be able to work around kangaroos and the likes of 

I'm going to have a lot of fun trolling the next generation.

My old GU Patrol might not be in that league yet but with no electronics and a good steel bulbar it should be around crushing cockroaches along with others of its vintage long after all the autonomous cars have been abandoned.

Edited 20September, 2017 by Redracn

[tazzieman]

My old GU Patrol might not be in that league yet but with no electronics and a good steel bulbar it should be around crushing cockroaches along with others of its vintage long after all the autonomous cars have been abandoned.

We'll be natting about the ol' OH&S days long after the sun's gone down and the whisky & soda no longer hangs off the front bumper.