[ecenandu]

Originally Posted by GTO

One thing we've missed out on is the cost. If & when self-driving cars do come, how much more will they cost over a regular car? 30 - 40%? How many will be willing to pay that kind of premium?

All future Tesla models will have level 5, fully autonomous capable hardware. Upgrade optional for the current fleet.

all Tesla vehicles produced in our factory – including Model 3 – will have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver. Eight surround cameras provide 360 degree visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength, capable of seeing through heavy rain, fog, dust and even the car ahead.
To make sense of all of this data, a new onboard computer with more than 40 times the computing power of the previous generation runs the new Tesla-developed neural net for vision, sonar and radar processing software. Together, this system provides a view of the world that a driver alone cannot access, seeing in every direction simultaneously and on wavelengths that go far beyond the human senses.

Source

[Jeroen]

Interesting thread!

Originally Posted by theMandarin

The referenced article addresses an interesting point that humans react based on common sense which is something difficult to program if not impossible.


The key takeaway for me was that none of the decisions were absolutely right/wrong, but merely had a varying degree of favorability.

What is called common sense in this context is merely a simple input for a computer. A traditional one is as follows:

You drive through your typical town urban type of street. Cars parked on both sides of the road and a sidewalk between the cars and the houses. Suddenly a tennis ball rolls into the road. Most drivers would brake or at least come of the accelerator. Not because the tennis ball is dangerous, but due to the fact that most of us are expecting a child to run into to road following the tennis ball. For a human we call this behaviour common sense. A computer just needs a good enough sensor to pick up that tennis ball and some logic to come to the same conclusion. Once that has been put together it will react much better and with a unbelievable high degree of accuracy and consistency to such a scenario. Much better than any human. A human could “miss” the tennis ball, or ignore the tennis ball.

Originally Posted by Nav-i-gator

One key difference with humans and "logic gates" driven cars - we can anticipate dangerous/potentially dangerous situations beforehand and this helps in increasing the time available to respond, while programmed machines would need a "stimulus", an input information in order to calculate and prepare an adequate response. Programmed self driven cars will do well (maybe even better) than humans in the event of an accident when it has just started (~0 second into a collision, when there is a need to limit the impact, prevent car from toppling, minimizing the drag, stabilizing the out of control car and so on) than preventing the accident altogether.

See above. Humans can anticipate but don’t always do. The response time of humans is actually pretty poor. See below.

Originally Posted by capslock

Sharing a web site related to this topic called The Moral Machine where a driver-less car must choose the lesser of two evils, such as killing two passengers or five pedestrians.

A lot of the debate around the pro/cons of fully autonomous cars seem to be centring around these sort of ‘moral’ dilemmas. To be honest, I don’t understand that, I also don’t think it is very realistic. As far as I’m aware there is no (scientific) evidence whatsoever that humans in such situation make this moral decisions appropriately. There are many very unfortunate accidents that show that human drivers just don’t make those decisions, freeze up, get it wrong etc.

For me the only criteria is actually whether from a statistical point of view an autonomous cars is much safer than the average human driver. The minute that comes true, putting autonomous cars on the roads will reduce road fatalities. That is just how it, statistically, works.

Of course, the moral dilemma is relevant for other reasons, such a (legal) responsibility. But if autonomous cars are positioned as being more safe, the moral dilemma don’t come into play as such.

Originally Posted by Whiplash7

Personally I don't favor the semi autonomous driving tech that is available in a few cars like the Tesla model S where you just sit in the drivers seat and watch the car drive autonomously but must be ready to jump in and take over. This reduces the drivers state of alertness and is difficult for the driver to instantly take over control unless he is fully aware of the current road situation up to the very last second.

I would say that all autonomous tech including level 5 should only be made to monitor the driver and take over when an accident is imminent, like a teacher watching over a student.

Actually, I think this is the most difficult part of autonomous driving. It is also the least understood by the general public. Although there is actually quite a lot of research done in the last decades in different industries. From nuclear power plants, to Industrial Automation and of course aviation. These are all industries that rely heavily on automated processes overseen by humans. And guess what, they all suffer from the same problem. When something unexpectedly happens that throws the automation out of whack, the humans are supposed to jump in and save the day. Fact is, humans don’t do that particularly well! Even those that have been trained professionally to a very high degree. Such as process operators, pilots etc.

Humans need considerable time to understand a particular situation and then again time to proces and to start physically responding to the situation by taking appropriate action.

So the idea that humans oversee a computer process and step in when the computer gets it wrong is actually not a very robust way of dealing with such a situation. I wrote about this when I visited Stanford University some time ago.
See http://www.team-bhp.com/forum/intern...omous-car.html

Took the average driver 3-5 seconds to respond when he/she had to take over from the computer. Even at a sedate 40 km/h a car would have travelled 30-50 meters before the human would start responding. Where does that put this so called “common sense” which some believe puts us apart from computers.

The solution? More automation, more computers, less human interference, not more.

Originally Posted by motorhead16

I don't understand why people think its so next to impossible to build autonomous cars in the near future.

It’s because there is no universally accepted definition on what constitutes an autonomous car. You could argue we have autonomous cars today. They are just not fully street legal, that’s all.

Originally Posted by noopster

I am very skeptical about the whole autonomous car concept. It's tough enough trusting another human being at the wheel, to put your life in the hands of a machine seems to be suicidal! It's not like autopilot mode in an aircraft which travels for thousands of miles of virtual solitude.

I don’t want to upset you even further, but planes don’t fly thousands of miles of virtual solitude on autopilot. They fly tightly packed together with relatively (to horizontal speed) very small separation and vertical separation is measured in hundreds of feet only.

Also, again not wanting to raise your anxiety about automation taking over the world any further, autopilots don’t have automatic collision avoidance built in. It always takes pilot intervention! Not a single commercial plane out there, in that very density populated air space moves away from a potential collision autonomously!

Commercial planes are kept separated thought Air Traffic Control and they use very sophisticated computers to track and monitor planes, weather, to forecast scenario’s etc. That is then communicated by ATC (humans) to pilots (humans). That is where it breaks down. Very few cases where the computers on the ground got it wrong. Plenty of cases where ATC-human gets it wrong, communication goes wrong or the pilot gets it wrong.

Remember aviation safety record has increased over the last couple of decades significantly to an almost unbelievable level. But the weakest link is still the pilots and even today in the vast majority of cases it is pilot error that causes (fatal) accidents.

Originally Posted by noopster

The other day I saw a car driving on the opposite side of the road IN REVERSE (I am not making this up). My brain could barely comprehend the situation let alone deal with it
Turns out the guy had smartly taken the wrong side to avoid taking a U-turn but realized halfway through that there was a cop stationed at the intersection where he was planning to merge into the correct side. So he reversed back to the next place where he could merge without getting a ticket.

Confused? Now imagine the state of the poor autonomous car!

Actually, your example is the best proof in favour of automation and autonomous cars. Indeed a human would be confused. But even for today’s autonomous cars this would be no problem. They don’t care whether a car is reversing or not. For them it is just an object in motion and they will respond to it with no hesitation in a very consistent way. Time after time! Computer 1, human 0!

Originally Posted by Rohan24

I personally feel Semi-autonomous tech is a better solution than complete autonomy. Complete autonomy gives the user *refrains from saying driver* a false sense of security and makes them complacent, and oblivious to whatever is going on. On the other hand, semi-autonomous tech needs the driver to be alert.

Yes, a lot of people think like that. Problem is, see my earlier comments. Humans such at stepping in and responding adequately. We like to think we are better, but there is no evidence that I’m familiar with that even remotely suggest humans are particularly good at this very task of stepping in, unexpectedly, into ambiguous situations and taking fast and adequate actions. To some extend it can be taught, e.g. rigorous continuous training and simulation. (e.g. pilots, process operators etc.)

Originally Posted by Rohan24

Usually, one can get bored on motorways, cruising at constant speeds, and its also tiring. Semi-auto mode will let you relax a bit in this section, so when you get to that twisty, driver's paradise, you can enjoy to the fullest.

Again, to my earlier points. This is the most dangerous scenario for semi-automous cars. With a 3-5 second response time at 100 km/h your car will have travelled 84-140 meters before the driver starts taking action.

Again, to me the debate needs to concentrate on what makes a safer car, not necessarily a better car, or a car with a better or equal moral compass as an average human.

That is why automation was introduced into process automation and the cockpit of planes. Yes, there are trade off, but the fact remains that overall safety increased significantly. Everybody in those industries is acutely aware that the weakest link actually is the human.

The fact that the autonomous car debate still centres around whether humans are better at driving then computers shows we still have a long way to go.

Jeroen

[nishsingh]

Originally Posted by GTO

While it's *at least* 10 years away in the USA,

Looks like Tesla desperately wants to refute that!

From Tesla Press Release yesterday (19th Oct)-

"We are excited to announce that, as of today, all Tesla vehicles produced in our factory – including Model 3 – will have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver. "

To do this, Tesla is using -
1. Eight surround cameras provide 360 degree visibility around the car at up to 250 meters of range.
2. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system.
3. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength, capable of seeing through heavy rain, fog, dust and even the car ahead.
4. And finally, a new onboard computer with more than 40 times the computing power of the previous generation that runs the new Tesla-developed neural net for vision, sonar and radar processing software.

Tesla claims that "this system provides a view of the world that a driver alone cannot access, seeing in every direction simultaneously and on wavelengths that go far beyond the human senses."

Looks like the technology is already here, just that we weren't prepared for it!

[Jeroen]

Originally Posted by nishsingh

Looks like the technology is already here, just that we weren't prepared for it!

As I said earlier, just depends on the definition (and therefore criteria) of an autonomous car. If safety is the key criteria, I’m sure this could be true!

Jeroen

[SmartCat]

Those looking at Tesla should remember that they are burning cash by billions. $4 billion in revenues and $890 million losses in 2015. What this means is - if they want just break even, they need to increase prices by 25% atleast. Also, progress made by Tesla in autonomous driving can be easily explained -> When you throw enough money behind something, you can achieve wondrous things. A good automotive example is Bugatti Veyron.


Anyway, Impressive demonstration video by Tesla - but Jalopnik is quite critical about this achievement
http://jalopnik.com/teslas-proof-vid...ver-1788018454

Tesla's Proof Video For Its Self-Driving Tech Is Some Fantasy Land Bullshit

The weather is crystal clear. The road is perfectly smooth. The lane lines are fresh and completely undisturbed. Highway merges are simple and signs are extraordinarily legible.

These sorts of conditions might exist in Palo Alto, where the video was shot, but the rest of the country has to deal with cracked pavement, potholes, blotchy lane lines if any at all, gravel, sleet, slush, rain, fog, confusing old merges, obscured signs, and basically just about everything that trips up current Teslas in autonomous mode. This video did nothing to show that Tesla had improved any of that.

[Acharya]

Electronics in todays modern cars continue to remain one of the top reasons for vehicle breakdown. ECU failure is not very uncommon, we heard the story of Q7 breakdown during Leh climb thanks to advance electronics.

With this background, how many of us are willing leave our lives in the hands of few sensors and computers, which can always malfunction outside their ideal conditions.

Even the airplanes, which are using auto pilot for decades, provide manual override option, as computers cannot react to uncertainties of life better than humans.

Thinking about cars without steering/pedals is just that - 'wishful thinking'.

[Nav-i-gator]

Originally Posted by Jeroen

Interesting thread!


For me the only criteria is actually whether from a statistical point of view an autonomous cars is much safer than the average human driver. The minute that comes true, putting autonomous cars on the roads will reduce road fatalities. That is just how it, statistically, works.

Of course, the moral dilemma is relevant for other reasons, such a (legal) responsibility. But if autonomous cars are positioned as being more safe, the moral dilemma don’t come into play as such.




Jeroen

Statistics are just that, what one make out of it is important. Statistically, IVF have better chances of getting pregnant than the "Natural" way. But that don't make us go directly to sleep in the night

Nobody is saying that autonomous cars are impossible, they are just far away to become a mass reality. More than the technological evolution, it is the moral, legal and behavioral changes that will take time.

The mention of most flight accidents happening due to pilot error is interesting. Here is my take to it - with so much advancements in technology to assist pilots, they are still trained to fly without any help from on-board computers, just in case the electronics fail. But pilots, over a period of their flying career, become so much reliant on electronics, avionics and controls, they forget to act as learnt when the need arise, and hence commit errors in judgement. In most of the cases, pilot error is preceded by, or accompanied by computer error.

in post-crash investigations (throughout the history) an error in construction, a faulty control switch or failure of sensors does not cause news (it is a mere stepping stone in further evolution of better designs and technologies), but a pilot error is followed by moral and legal outcomes.

A whole new set of laws and legislation has to be made to enable autonomous cars on road. Who will face the court in case two autonomous cars are involved in a crash?

More pertinent scenario - A lot of accidents usually happen when the driver, in order to avoid hitting a stray pedestrian/biker on road, swerves sideways, lose control and crash, many a time fatal to the car occupants. How will a car be programmed to "judge" whether to hit the pedestrian/biker to avoid injuries to the car occupants or vice versa?

It is not the speed of judgement, it is the morality involved in judgement that is a hard nut to crack.

[ecenandu]

Originally Posted by Nav-i-gator

It is not the speed of judgement, it is the morality involved in judgement that is a hard nut to crack.

As Jeroen has pointed out earlier, will a human counter part take an ethical choice in those situation? Given the time required to react is so short, it will most likely be a knee jerk reaction.

We don't judge the morality of the driver in those situations right, that is why we call them accidents.

Anyway, I clearly understand your point of view about the morality in case of autonomous cars. Here is my solution for that, every car should have a blackbox equivalent data recording system and fatal crashes should be investigated like air crashes. This is mainly to find bugs/faults in the software or hardware of the autonomous car.

I don't understand the reservation of people in trusting 'few sensors or computers'. In fact, most of the modern cars have plenty of them for engine management, airbags, ESP, ABS etc.

The technology involved in autonomous driving wasn't a quantum jump actually.

Cruise control -> Adaptive cruise control -> Lane keep assist -> Tesla auto pilot 1. See, you get the picture right, you can see gradual increase in the sophistication of the technology involved in the above systems.

The autonomous driving software is totally different from the traditional softwares we currently have, mostly they are not rule based at least the system from comma ai isn't. They are trying to mimic the way in which a human drive.

I have posted few videos about comma ai in another thread, quite informative.

[Nav-i-gator]

Originally Posted by ecenandu

As Jeroen has pointed out earlier, will a human counter part take an ethical choice in those situation? Given the time required to react is so short, it will most likely be a knee jerk reaction.



I don't understand the reservation of people in trusting 'few sensors or computers'. In fact, most of the modern cars have plenty of them for engine management, airbags, ESP, ABS etc.

I never said that autonomous cars are an impossible future. Just said that more then tech advancements, it is the moral and ethical environment that need to be developed first before the autonomous cars are allowed to ply on roads

Maybe the drivers react to any accident in a "knee jerk" reaction. But programmed cars can't do that. An autonomous car will react as per the algorithm, and it is very difficult to program an algorithm for ethical choices when there are lives at stake.

It reminds me of the movie I-robot where the AI robot rescued the drowning will smith and let his daughter die (as the chances of survival were more for will smith than his daughter - calculated by the programmed robot). The hero (or in reality, if encountering such situation) a man would invariably try to rescue his daughter, even if he know he wont be able to, or he may also die doing so.

Some real life situations, and the way humans react to them instantly, may seem to be knee-jerk, but probably are well calculated in our sub-conscious mind.

[ecenandu]

Originally Posted by Nav-i-gator

...

it is the moral and ethical environment that need to be developed first before the autonomous cars are allowed to ply on roads

We should have legislations which make sure that level 5 autonomous driving will be done only if we have sufficient data to prove the safety. Similar to what we have in aviation industry.

That is the main reason why I said we should have black box equivalent data logger on every autonomous cars. Since there are no perfect technologies, there will be always room for improvement.

Maybe the drivers react to any accident in a "knee jerk" reaction. But programmed cars can't do that. An autonomous car will react as per the algorithm, and it is very difficult to program an algorithm for ethical choices when there are lives at stake.'

Don't underestimate these algorithms, there has been a lot of development in this field like fuzzy logic, neural networks etc. So, we cannot say it is impossible to program software for ethical choices.

The hero (or in reality, if encountering such situation) a man would invariably try to rescue his daughter, even if he know he wont be able to, or he may also die doing so.

Not every person could do that right.

Some real life situations, and the way humans react to them instantly, may seem to be knee-jerk, but probably are well calculated in our sub-conscious mind.

The biggest advantage I can see with autonomous driving is that these systems won't text, sleep or be intoxicated while driving. And in those real life situations, where a human has a split second to decide, the system will have far more data, like 360 degree view of its surroundings, radar data etc to make a well calculated decision than a human.

Incase the system make an irrational choice in certain situations we can improve the software, right.

[Nav-i-gator]

Originally Posted by ecenandu

We should have legislations which make sure that level 5 autonomous driving will be done only if we have sufficient data to prove the safety. Similar to what we have in aviation industry.

That is the main reason why I said we should have black box equivalent data logger on every autonomous cars. Since there are no perfect technologies, there will be always room for improvement.



Don't underestimate these algorithms, there has been a lot of development in this field like fuzzy logic, neural networks etc. So, we cannot say it is impossible to program software for ethical choices.

The hero (or in reality, if encountering such situation) a man would invariably try to rescue his daughter, even if he know he wont be able to, or he may also die doing so.

Not every person could do that right.


The biggest advantage I can see with autonomous driving is that these systems won't text, sleep or be intoxicated while driving. And in those real life situations, where a human has a split second to decide, the system will have far more data, like 360 degree view of its surroundings, radar data etc to make a well calculated decision than a human.

Incase the system make an irrational choice in certain situations we can improve the software, right.

Again, I (as well as the thread title) am not saying it is "impossible", merely pointing out the real life issues to be settled, rather than the techlology itself.

As of now, the most advanced autonomous technologies are used in space travel, where satellites and probes are programmed to run certain parameters and autonomously perform when it become difficult to control them in real time owing to their vast distance and time taken to communicate. But they are also programmed only for "known variables", not the unknown ones. a probe can autonomously carry out a slingshot fly-by based on pre-fed instructions of speed, trajectory and precise engine burns (all known variables) but as of now, we are way behind the day when we can just blast off a probe with all the known tech like fuzzy logic, neural network etc from a rocket with an instruction to just "go and find your way to pluto yourself".

Such time is "l-o-n-g time away".

Coming to the isuue of moral/ethical dilemma - to put it as simple as possible:

I would rather make my own informed choice in a life-and-death situation and agonize over it if it turns out to be wrong than allowing someone else to decide my fate (much less to a program) and spending my life with remorse as to maybe I would have done that in a different way had I done it myself.

The day machines become so advanced and sentient, we can think about that then.

[ecenandu]

Originally Posted by Nav-i-gator

I would rather make my own informed choice in a life-and-death situation and agonize over it if it turns out to be wrong than allowing someone else to decide my fate (much less to a program) and spending my life with remorse as to maybe I would have done that in a different way had I done it myself.
.

Absolutely your choice buddy, I appreciate that. Does this mean that even if there is a well proven level 5 autonomous driving available, you won't use it?

EDIT:

The day machines become so advanced and sentient, we can think about that then.

Okay, that means you will wait until it is advanced enough to handle those situations. So will you skip the intermediate levels ?

[Nav-i-gator]

Originally Posted by ecenandu

Absolutely your choice buddy, I appreciate that. Does this mean that even if there is a well proven level 5 autonomous driving available, you won't use it?

EDIT:

Okay, that means you will wait until it is advanced enough to handle those situations. So will you skip the intermediate levels ?

If there is no option of a manual override, I would have my reservations using it (there is no point in not baing able to use a car just because something as mundane as a GPS reciever/ a camera/a radar is not working). or atleast not till the "rise of machines"

[silverado]

Fully autonomous cars are a long time away - From being commonly used.
Fully autonomous cars will at some point become common on roads, though I don't think they will come to india.

I feel the technology can be used in many urban scenarios such as bumper to bumper traffic even in our road conditions.

Autonomous cars if limited to slow speeds ( below 10kph ), can reduce a lot of stress from driver in many situations. Since the speed is slow the risk involved would be low.( Assume cost will also be lower )

So if we get a car which can drive autonomously in bumper to bumper traffic and costs 10-15 % more, it will be a boon to use.

[Samurai]

http://money.cnn.com/video/technolog...inkId=30320745