Slashdot Mirror
GM Exec Says Elon Musk's Self-Driving Car Claims Are 'Full of Crap' (smh.com.au)
An anonymous reader quotes the Sydney Morning Herald:
Billionaire entrepreneur Elon Musk's claims about the self-driving capabilities of his upcoming Tesla vehicles are "full of crap", General Motors' self-driving Tsar says... "To think you can see everything you need for a level five autonomous car [full self-driving] with cameras and radar, I don't know how you do that"... GM's own solution involves several radar and Lidar sensors, as well as cameras and multiple redundancy systems. Each system costs hundreds of thousands of dollars, and GM are some way away from getting the cost low enough to be commercially viable. "The level of technology and knowing what it takes to do the mission, to say you can be a full level five with just cameras and radars is not physically possible," Mr Miller said.
GM can't do it.
I definitely trust GM to make an unbiased analysis of competitor technological capacity.
I don't possess radar, LiDAR, or a gazillion redundancy systems. Stereo cameras (eyes) on a pivoting head and two directional microphones (ears). My software is way better than GM's, though, and I'm expecting Tesla's is too.
Then why can't computers. We have an advantage that we can fill missing pieces in stereoscopic vision to complete our perception and it is incredibly difficult for algorithms to do that. Radar makes up for that, but it is simply possible that Tesla is closer to that then GM is.
I function in level 5 autonomy without lidar and radar.
Our current primary system for operating cars relies on 2 cameras and no radar/lidar. People have a fraction of the sensors that even a tesla has at this point. Sure the programming is hard and the compute on current vehicles might not be sufficient but I think Musks approach is more likely to succeed. You work on the hard software problem and find a solution instead of just throwing more hardware at the problem and hoping you can eventually get the cost of the hardware down enough for it to actually be applicable.
Then so can a computer. Just need the right computer and software behind the cameras.
Mike @ The Geek Pub. Let's Make Stuff!
But to clarify the difference:
LIDAR: Formerly about $75k, now about $7,5k per unit, and requires a bubble dome on top of the vehicle. GM and Waymo use it, Tesla doesn't. In addition to looking weird and adding drag, the price is killer if you want to include something on every vehicle. Beyond this, LIDAR doesn't work in fog, heavy rain, snow, and other conditions that humans can drive in - meaning that you'd have to either prevent trips during these conditions, require humans to drive during them, find workarounds (not easy), or rely instead on other sensors. And you still need to understand the world around you visually - LIDAR will tell you that "something" is there, but it can't read signs, see road markings, see brake lights, tell if that thing in the road is a person or a paper bag, etc.
Tesla, for these reasons, ruled out LIDAR. They just simply use the "other sensors" - 1x radar, many cameras, many ultrasonic sensors - all the time. This way, all of their sensors can be put in all of their vehicles, and do double duty for both self driving and standard safety features (autobraking, etc), depending on what options the buyer has paid for. This however comes at a penality: when LIDAR works, it works really well. Photogrammetry with cameras is prone to stitching errors, and radar, while being able to see some things that humans can't, sees the world in very strange ways (for example, a piece of plywood is transparent, but an aluminum can glows like it's on fire). It's a much more challenging task if you leave LIDAR out of the loop. But, it gives you a more saleable product.
In the end, I expect a convergence to take hold. An interesting new technology for example is time-of-flight cameras - they function as normal cameras, but also can read the length of time it takes for a laser pulse to return on every pixel they record. So no dome, just your normal camera coverage and a few cheap, fixed lasers - in mass production, it might not cost much more than cameras alone. In such a case, I'd expect the LIDAR groups to simply replace their conventional LIDAR datastream with the time-of-flight datastreams, while I'd expect the non-LIDAR groups to replace their photogrammetry-and-radar built 3d models with time-of-flight 3d models. But both sides will still need image processing, so it's important to work on maturing that technology today.
That said, let me reiterate that I'm a pessimist regardless of what tech you use. There's just so much nuance in driving in hazardous conditions - understanding when, where and how much you have to slow down, what's safe to drive on and what's not, what things to the side of the road are hazardous and what aren't, when you should break rules (such as driving in the middle of the road when conditions are dangerous but oncoming traffic is rare), what are the consequences of a mistake in one location vs. another, etc, etc. On my gravel road, there's a canyon to one side with no guardrail, and varying amounts of ice and potholes in different places. You better well know how your traction is going to fare as you move across the potholes (vibrating the car and making it lose traction) or icing if you don't want to end up in an unrecoverable slide into a ravine.
Just to pick a random example among countless things that you have to take into account: how long do you think before any self-driving systems will have "sheep recognizing algorithms"? Because where I live, there's sheep. Group of sheep on one side of the road: probably safe. Group of sheep on both sides of the road: not as safe, but probably safe. Lamb on one side, ewe on the other? Very dangerous - the lamb will invariably run to its mother as you approach. Where's the ewe-lamb-running algorithm?
"If there was an antonym to 'Elon Musk', it would be 'Richard Branson'."
With a hammer, everything looks like a nail. Turns out, some things are bolts.
Theorically, only two cameras, a few accelerometers and microphones should be sufficient, because humans don't have anything more.
Practically, as the goal of autonomous vehicle is being better than humans, more sensors will be certainly needed.
There is also the cost of redundancy. You need to duplicate (or triplicate) everything for true fail-safe systems.
The practices already used in "safety critical" industries will be applied on autonomous road vehicles, this is how aircrafts became reliable.
Also, as to answer the question of how humans do it with "two cameras": logic. We don't have "stitching errors" in how we build up a model of the world around us from visual data because our brain constantly processes everything around us through the prism of "does that make sense?" But whether something "makes sense" or not is an AI-hard problem.
Building up 3d models with photogrammetry is an inherently error-prone process because a computer doesn't know if something makes sense. The approach is "Oh hey, these patterns from the left and right camera matched up - there's an object there at X distance based on how far the patterns had to be shifted to align". But what if the patterns happen to be different things that just happened to match up in patterning? Or what if, due to lighting / texturing / obstruction / material issues, the same thing didn't look exactly the same from different angles? Uncovering these problems is, as mentioned, AI-hard. I doubt anyone is even trying at this point; there's enough to work on just to get things to follow road lines correctly and not go chasing old tire tracks or poorly erased construction lines.
Ranging systems like LIDAR help let you just simply ignore the issue by telling you flat out, "I sent out a beam in this precise direction and got a reflection in precisely this length of time." But LIDAR has a number of problems, as described above.
"If there was an antonym to 'Elon Musk', it would be 'Richard Branson'."
Also, we crash more often than is permissible to a self-driving vehicle.
What GM is saying is clearly false.
People can already drive and they only have two eyes and limited mobility to evaluate the outside environment.
It is a matter of computing power only, which some compensate for by adding more sensors, but I wouldn't go as far as saying that there is a clear recipe for an autonomous car that everyone has to follow.
The statement is meaningless unless the environment is qualified.
Most anybody can do it on a controlled access road.
It may not be "perfect", but it will be better than you.
How do you know that is a dangerous situation ? You didn't get born with it. You know that is a dangerous situation because you learned about it or experienced that behavior. Just like an AI would have to.
City folk will still drive right through, unless they brake to watch the cuteness. So it is still only a matter of training the AI.
Because where I live, there's sheep. Group of sheep on one side of the road: probably safe. Group of sheep on both sides of the road: not as safe, but probably safe. Lamb on one side, ewe on the other? Very dangerous - the lamb will invariably run to its mother as you approach. Where's the ewe-lamb-running algorithm?
....or Tesla, or any American car manufacturer.
People in the know all look to the European makers, with the Japanese/Koreans possible seconds.
Americans are incapable of making decent normal cars, let alone self driving ones. The Europeans will again, push the envelope. This technology will gradually trickle down to the more mundane car manufacturers, just like seatbelts, windscreen wipers, disk brakes, ABS and traction control did.
It's really remarkable that the Americans can produce a technological marvel like the SR-71, yet can't make a car worth a shit.
The self-driving car had better crash less often or why am I spending 10 of thousand more dollars on it.
It's funny that "I don't know" somehow gets translated into "it's impossible!" I recall a few fools claiming they had invented a new "unbreakable" encryption scheme because they didn't know how to break it.
Anons need not reply. Questions end with a question mark.
titles
Self-driving systems are not "generalized AI"; they don't learn from the ground up. You might incorporate neural nets in specific image recognition tasks, but in general, self-driving algorithms will only ever do precisely what you tell them to do - and that's for good reason. Not just because of the limited capabilities of today's neural nets, but also the simple fact that if something goes wrong, you want to be able to say "Here's what went wrong" rather than "Oh, dang, this black box decided to drive this person off a cliff for some inexplicable reason. Oh well..."
"If there was an antonym to 'Elon Musk', it would be 'Richard Branson'."
In particular when you take into account the other insane claims Musk makes all the time. There is a certain (unfortunately large) audience that cannot distinguish fantasy and reality, and Musk caters to them with great success. The one thing I am undecided about is whether he himself is just artfully manipulating the morons or whether he also has trouble separating fantasy and reality. At this time, both options are credible.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
We need low-cost electric cars for the common man, not expensive sort-of-self-driving electric cars for the rich.
#DeleteFacebook
Next you're gonna be telling me how we're not all going to have villas on Mars by the end of the next decade, at the very latest, all brought to you by SpaceX brand rocket ships.
Elon gets my respect for making two successful and innovating businesses that have lasted and have solid fundamentals into the future. He gets no credit for his bullshit factory. Good bullshit has to be believable. Self-driving cars and 800 mph trains in a tube by next year doesn't pass the giggle test.
I've driven in a vehicle that has been out in a snowstorm. I'm wondering if there will ever be a sensor array that will be sufficient for automatic driving without having to spend three hours cleaning snow off. Generally you bear 10-15 minutes of cleaning, but you're not getting every square inch of every window. Airplanes where I am get hosed off with antifreeze before every flight; not very realistic for automated car owners.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
I am pretty sure the machine learning robots will know about sheep after ingesting billions of hours of Russian dash cam footage
That's easy. Assume all of those ewes and lambs will try to throw themselves at your vehicle, calculate how far they can get before you can come to a complete stop, and slow down accordingly. This algorithm also works for children playing by the side of the road.
Any sufficiently unpopular but cohesive argument is indistinguishable from trolling.
I'm not pessimistic. I absolutely believe a aging legislature will make it reality when they see it as a necessity for their freedom and mobility. Even if it fails sometimes, the times it stops a 80 y/o from backing up into a pedestrian and jamming the gas because something felt off will overrule it.
Didn’t a GM exec in 1970s say something about Japanese carmakers will fail in America because Americans didn’t want small fuel efficient cars?
Tesla must be quite confident they can make it work, as they are already selling cars with full self driving as an option. The web site says it will come as a software update one day, exactly when depending on regulatory acceptance.
const int one = 65536; (Silvermoon, Texture.cs)
SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
Silly GM didn't you realize you became absolutely irrelevant like a decade ago.
GM can't do it.
...so they did the next best thing which is to launch whiny ad hominem attacks on Tesla and Musk.
I think this all boils down to terms of service. I suspect that Mr Musk's terms of service are so "bad" that they can claim anything while a "real" car company like GM is really trying to solve the problem not just be first to market with a beta product. After the many lawsuits against GM over many years, I think their culture is very different in Detroit than Tesla's.
Self driving cars are the SAME kind of well organised fake news as 'theories' of 'race' differences in Humans (actually pseudo-science originated in early 19th century America in an attempt to justify Human slavery when the Old Testament teachings were no longer cutting it in US courts), and the so-called 'hysteria' illnesses of Human females.
What Google calls 'self driving' cars (on ordinary roads- on purpose made tracks self-driving has been a thing for many many decades) are actually DRONE cars- with remote operators who take over the control of the vehicle whenever conditions are 'difficult'. When a real car company falls for the hype and tries their own design (like GM), they can't understand why the problem is impossible to solve.
Google is pushing the concept to GROOM the general population to accept semi-autonomous ground travelling killing machines- the drone tank army Google is currently designing, in a TRILLION dollar project, for the US army. The grooming is similar to that which happened just before WW1 get get the sheeple to accept war from the air- the use of the newly invented aircraft to carry and drop bombs. Originally people considered such a thing demonic and utterly evil. The 'googles' of the time had to spend billions in PR to win the public around.
Now DRONE vehicles, with one operator per, say TEN vehicles are possible on good roads on certain chosen routes- but the cost factor is astonishingly hopeless given how much real drivers cost. And the accident rates of drone vehicles are such that eventually the government won't be able to hide them anymore, raising the insurmountable problem of insurance.
But, like I said, the idea is simply to get a new generation of engineers with the skill set to work on MILITARY drone vehicles, that can happily roll over a schoolbus full of kids in a nation where the press of the West tells you the people are 'sub-Human' and thus can be murdered at will.
Here's a thought for the hard-of-thinking (most people who still choose to come to slashdot). If civilian drone vehicles were a thing (and they are not in a real sense), nations like China would be building their new cities with futuristic road tech designed to make the use of drone vehicles as easy/safe as possible. Sensors bult into the road surface. Exact junction design. Perfect machine vision signage. But no where do we see these new types of roads built. Why? Cos civilian drone vehicles make no commercial sense at all. People like driving. People driven vehicles are good for even the most 'difficult' types of existing road design. People drivers are so much cheaper.
All this shit about sensors and shit seems overly complicated. It seems like you could combine aspects of a rail system with that of a car: an electric car that attaches to an electrified rail and follows it until the driver wants to get off the highway. It would probably be possible to insert a data/sensor system in the rail itself that tracks position and movement of cars ahead and behind so that they aren't too close when someone leaves/joins. (There might a problem if someone enters the lane with an old car, but maybe underground sensors could detect them?) There would still need to be an override in case deer or something get on the road.
Hell, I think projects like that California high speed rail should take cars on trains instead of just passengers. More people would use it if they could take their car along for a trip. In far-off sci-fi tech, each automobile/passenger group would have its own train car which would move to the back of the line and detach for their stop, rather than the whole train stopping. It might be easiest to have the exiting unit/car move to the opposite rail/guides (the main train cars would then slow down or speed up to rejoin), but I'd think you'd need a maglev for it.
Won't work until the car has good recognition on what's mobile and what's not, age of child and whether the child is accompanied by an adult (aka likelihood of being near the road and suddenly entering it). Otherwise even every tree/bush/sign/bike could jump into the road and the car will slow to walking speed all the time.
I think the human is going to have 'intuition' that will not be found in the first few versions of the machine (although the reverse will be true also). I think the machine is going to have much better reaction to a well formed obstacle suddenly entering the roadway (although for small items at high speed it likely can't try to avoid them any more than a human could).
I think the real mistake is not restricting autonomy to the Interstates and some well-maintained state highways (and safe places to park when exiting same) for the first couple years. They are limited enough that: the car can know exactly how many lanes and where they are, the car can probably get daily updates road construction and lane shifts and get by-the-minute or even by-the-second updates on traffic conditions... and especially have a little subroutine that compares what the car would have done if it didn't know in advance of something. (For example, the lane tracking routines would have gone straight, but the car happened to know a lane shift was occurring, or there happened to be another car ahead that it decided to mimic - if those with- and without- special help decision trees produce different results, the car maker needs to know that without special help the car would have done the wrong thing and in a different situation could have caused a fatal crash.) I have a funny feeling that upon getting 10^3 or higher number of would-have-made-the-wrong-choice reports from a well known roadways, the carmakers would thank their lucky stars for being able to improve their road/decision analysis without it coming from a fatality. And of course, the interstates always separate opposing traffic, and limit chances for objects, animals, and people to appear from nowhere.
This would also have had the benefit of still getting most of the benefit of taking over the boring part of long drives, and assuming that non-autonomous brake assist would still have avoided a large share of crashes in other situations. Unfortunately what we have now is blind hope placed in an industry that can't program well enough to not have OS patches each and every month.
Lidar was doomed from the start. If a car is going to be autonomous, it must function when drivers aren't paying attention to conditions. Otherwise, what's the point? Other systems will have to be good enough to work in fog. And if you have systems that can work even in poor conditions, then lidar is uselessly redundant.
Or do you too find it curious that they are measuring vehicle autonomy the same way we did VR? Remember VR-5? -T
Also, as to answer the question of how humans do it with "two cameras": logic. We don't have "stitching errors" in how we build up a model of the world around us from visual data because our brain constantly processes everything around us through the prism of "does that make sense?" But whether something "makes sense" or not is an AI-hard problem.
You'll find a million human visual logic errors on google images under "optical illusions." Was the dress white and gold or blue and black? Then there's the hot road mirage (https://www.youtube.com/watch?v=_M0FcpQWh5E). My broader point is that people don't see perfectly either, but if machines always drive sober, without texting, and not sleepy, they could eventually do better than the very low bar we are setting.
Multiple redundancy systems are not there for the tires or the brakes or the seat belts.
Why multiple redundancy be necessary for a self-driving system that can simply "bing bing bing" and demand the human "driver" take over?
Why multiple redundancy be necessary for a system that can simply pull to the side of the road, as a human would if they were to have an emergency?
This guy sounds like he comes from the defense industry. Multiple redundancy is nice in an attack helicopter that is 45 minutes behind enemy lines, but not necessary to take me 3 miles to work.
That is not going to be done by an algorithms but trained neural networks.
It seems there are two schools of thought here.
A: Drive better than a human in as many conditions as possible using weak AI and sensors that are fundamentally different than human senses (LIDAR and others).
B: Or drive like the most experienced human on the planet, using senses that are similar to what we use, (RADAR, spacial sound perception, stereo cameras), and a massive store of intuitive knowledge of how you've driven that very same road many times before, along with enough common sense to slow down or stop and wait for conditions to change for the better, or just call for help.
I can see where LIDAR would be great for exploring a new planet, but for good old Earth, I think most of us will accept getting a ride to the grocery store by our one eyed, half deaf grandparent who just happens to have been driving the same road for over 60 years. It seems to me that a weak AI with the equivalent experience of hundreds of thousands of drivers over the entire planet might be able to pass the "good enough" test very soon, if not already.
Elon's had a long history of proving naysayers like this wrong. My money's (literally) on him to pull this off. The folks truly terrified of self-driving cars are the National Auto Dealers Association (NADA) who stand to lose the most when you can order a car online and have it deliver itself.
Humans drive with real intelligence with only two eyes for input.
AI should be able to replicate the feat with only two cameras.
I bet google gets there first with the lead they have in object detection and model improvements. They also have the largest training dataset to work with--another key advantage.
"The level of technology and knowing what it takes to do the mission, to say you can be a full level five with just cameras and radars is not physically possible,"
Not on Von Neumann based architectures.
http://memcpu.com/
Things are changing, old timers.
During 2015 there were an average of 88 fatal car crashes a DAY in which an average of 96 people a DAY were killed. Can AI do any better? More than likely, but time will tell.
The BIG problem is that cars do not communicate with each other and so they cannot coordinate their movements. Being able to do that would reduce the causes of accidents and fatalities to just mechanical failures. Sensors and communication networks buried in roadways would help significantly. IF cars were networked, for example, they could all accelerate from a stop light at the same instant, instead of waiting for the car ahead to begin moving. Cars could queue in and out of a line of cars automatically without having to wait on someone being "courteous". No more T-Bones, head-on crashes, tail-enders, speeding, or driving too slow.
Since cars are primarily meant to take a person from point A to point B they can become a standardized public utility, releasing many people from an obligation to purchase a personal vehicle, maintain it, service it, insure it and store it. Just like a cab today, but a LOT cheaper and with an AI driver.
Oh, the GM guy explained why GM and the rest ceased to be market leaders decades ago: " I don't know how you do that". That much is obvious.
Running with Linux for over 20 years!
In fact, I know personally of a human with just one eye. He drives just fine. So, one camera should be sufficient.
Well said. I enjoy driving my "Autopilot" Tesla on interstate highways with me paying attention. I just don't see Tesla or GM getting to Level 5 and not having a human available for backup. I think autonomous flying would be an easier thing to develop but I don't see myself getting in to a 300 passenger plane and flying across the Pacific without a trained human pilot able to take over.
Power tends to corrupt, and absolute power corrupts absolutely.
...is what happens when this tech is on every car? It's all very well to test and develop these things in isolation in California, Nevada or Arizona during bright sunlight.
What about forty or fifty vehicles at a busy intersection, all firing ultrasound, LIDAR and/or microwave in every direction, at night, in the rain? The scope for false positives and false negatives is immense.
Or perhaps the makers will modulate all the output with a unique identifier, perhaps the VIN. So then what happens to your privacy when your identity is being broadcast to all and sundry? Further, what happens when someone spoofs your supposedly-unique output? At high speed and in heavy traffic?
Anyone who thinks that true self-driving vehicles will exist in quantity in the real world beyond specifically-controlled niche use-cases any time soon is delusional.
This sig left unintentionally blank.
Computers have no intelligence.
You can program them to simulate the effects of having intelligence, but you can not make them intelligent.
No computer "knows" anything. No computer actually understands anything. You can give a computer the equivalent of a massive cross-indexed dictionary and code that can ge given a word like "car" and look up information about the car to present to a user - but the computer itself will not actually know and understand anything about the subject or even about what it is doing. Computers are not self-aware and cannot be made self-aware - and writing a program that claims to be self-aware is NOT the same thing as a machine actually being self-aware. Things like the Turing test are not valid either - the Turing test really just tests whether some code is sufficently complex to fool a human, NOT whether there is any actual computer intelligence being displayed.
Humans can look at a scene with their binocular vision and immediately recognize everything in the scene and judge distances, speeds, etc based on a knowledge of all the stuff in the scene. Computers cannot do this because they do not acually KNOW anything. You can give them a library of things they can simulate recognizing, but they do not actually recognize anything and understand what it is (they just do a version of pattern matching) so when they encounter something unknown in the scene before them bad thngs can happen. The problem is massively complex and if any situation arises which the programmers did not properly anticipate, people can end up dead.
Oh, and another important note: A human driver can lose sight in one eye and go right on safely driving without the benefit of binocular vision because he is intelligent and recognizes everything in the scene and can use knowledge about the items he sees to supplement the 2D visual info.
None of this means self-driving cars are impossible, and I am certainly no luddite. When the tech gets to a low enough error rate that the crashes are no more frequent than human driver caused crashes, we should switch to it. We should all, however, have a much more realistic view of the differences between the actual intelligence of an actual living creature and the simulated intelligence provided by tons of code pushing bits through millions of transistors.
You make very well-thought points about the technical stuff, but I disagree with your conclusions. I don't think your pessimism is logical based on the points you've made.
Group of sheep on both sides of the road: not as safe, but probably safe. Lamb on one side, ewe on the other? Very dangerous - the lamb will invariably run to its mother as you approach. Where's the ewe-lamb-running algorithm?
Sure, but to be fair if I had to travel to your area and rent a car, not knowing anything about the behavior of sheep, as a human driver I would be pretty terrible compared to you in that situation. But I doubt you'd say I don't deserve to have a driver's license, because that's a pretty unusual circumstance. I also doubt that even in your area, when you were taking the driving written test, that they had a section on whether you understood the intricacies of sheep behavior.
That's not because your experiences aren't valuable in that situation: it's because there's a difference between a driver and an experienced driver. You get the basic rules of the road going, and statistically you're safe enough to be trusted to be on the road. Young drivers are statistically less safe than experienced ones though. And you will tend to gather experience that relates to the particular location you drive in. Case in point, I'm pretty good at anticipating behavior from drivers in my area, but I've driven in another country and the same cues didn't apply. Still, I didn't get into an accident, being less experienced increases your risk, but it can still be an increased yet acceptable risk.
Where am I going with this? Well, autonomous driving is like an inexperienced driver with a lot of raw talent. It has an advantage over every human driver, including you, in that it will **never** get distracted, or tired. It has a field of vision of 360 degrees at all times, instead of having to choose what it will look at now. It has sensors that can see through objects you can't, so when I'm trying to turn in an area where the parked cars obstruct my view of an incoming car? It doesn't have that problem, it sees the incoming car.
Will it make worse decisions than I would given all the same information? During unexpected hazardous conditions, and assuming I'm not driving after not getting much sleep last night because I had to do another all-nighter at work...probably so. But the fact it has more information than I do and invariable alertness means that even though it would get me into accidents I wouldn't get into if I were driving, it nevertheless will get into less accidents, because it will avoid many of those more common accidents that I would get into if I were driving.
Not only that, but the sheep avoidance algorithm? It's coming. We're talking about the early inexperienced autonomous vehicle, but it's not always going to remain that way. Every time there's an accident with an autonomous vehicles there are logs to be examined. Engineers will pour over them and try to figure out if there's anything that can be done to avoid that type of accident. The problems that occur most often are fixed first, but eventually the system will be good enough that anticipation of animal behavior will be significant. At which point an update gets pushed to your car, and it now knows to slow down and be ready to avoid that road-crossing lamb. Because, unlike humans, when *one* autonomous vehicle gains experience, *all* autonomous vehicles gain that experience. And that's reason to be pretty optimistic.
Warning: Opinions known to be heavily biased.
Obadiah Stane: Tony Stark was able to build this in a cave! With a box of scraps!
It's hard
I'm skeptical that fully autonomous vehicles will be perfected any time soon
As optical illusions prove, inferring 3D structure from stereoscopic input is also error prone for humans. I guess that makes it I-hard?
I think GM is heavily invested in delaying self-driving as long as possible, mainly by making it as expensive as possible. Once fleets of self-driving cars become available, people will be buying FAR fewer cars. In twenty years car ownership will seem quaint and driving will be a lost skill like cursive writing.
I don't think that follows. We do not know how people think, yet people can tell us why they do the stuff they do. An AI can certainly do the same thing, it's just another output. Plus, you can always log all the actions and inputs anyway.
Autonomous driving has to prove it can deal with more than driving, it also has to be relatively capable of dealing with human beings trying to fool it into doing something else - hacking of a sort. If someone can stand on bridge over the 10 during rush hour, then wave the right poster to trip the right zero-day, that would be a spectacular terrorist attack.
Between how GM does things and how a startup can do things. So I call the head of GM an idiot for this one. It's funny I'm in the market for a new car - saw a review of the Chevy Bolt - I applaud that they have an available electric car, but the interior on the car is as cheap looking as you can get.
AI can drive better than humans
Its simply a fact with the available numbers.
What level am I, GM?
Many optical illusions happen exactly because we build up a model of the world based on what makes sense to our brain based on past experiences, and carefully-constructed drawings can exploit the assumptions made by those models.
I also remember when ULA insisted that reusable boosters would never work right or be reliable enough for a second flight.
That's our life, the big wheel of shit. - The Fat Man, Blue Tango Salvage
That’s the Old GM you’re thinking about. The Old GM that went bankrupt about 10 years ago. This New GM (actually a completely new company on paper) is slightly more relevant, they have actual multiple EV whereas the Old GM killed their EV1. I’m all for some competition in the car industry. I wanted to get a Volt (Bolt hadn’t come out yet) when I bought my last car but the economics didn’t make sense over a cheap Prius ($22k vs $37k). Would have never made the money back in gas savings.
As long as overall crash statistics improve the autonomous system will be considered a win. It's pointless to keep bringing up scenarios (sheep, children etc) where emerging AI may still fail compared to human logic. Of course over time the goal is to improve on all aspects but for now let's just worry about 800-some car accidents per day in Illinois alone.
...he could have responded with "He says he has Level 5 AI with radar and cameras? How many people have died in Teslas relying on their AI, again?"
-Styopa
GM guy got his panties wet over yet another V6, front-wheel-drive, 4-door sedan. So original.
In a world of the blind, the one-eyed man is king--and the two-eyed man is a heretic.
Says the leader of fucking GM. When was the last time anyone said "yeah I'm excited to buy a fucking GM car"? Probably somewhere near 1981 when technology meant plush seats, and adding a Pontiac symbol (a now defunct company!) to some shitty battletank of a car. Fuck GM and their obsolescent, shitty, shallow opinions on the technology industry. Let me know when they are successful in finance, solar, and automotive industries before making comments.
GM can't do it, so nobody can. What a douche.
The GM CEO view is a very common one: You can only get perfect data from perfect sensors, and Tesla's sensors are far from perfect.
That view is common,but it is incorrect from two separate perspectives:
1) There is no such thing as a perfect sensor. At best there's "close enough", defined as the point at which more money does no good.
2) The use of multiple sensors, and multiple kinds of sensors, creates a whole that is far greater than the sum of its imperfect parts.
The problem then becomes not just getting this data from one special sensor, but getting all the data we need from a set of imperfect sensors.
This is called "sensor fusion". At a simple level, it makes sense that two cheap radar sensors ought to be able to create better data than a single sensor costing twice as much, because each of the two sensors can be used to check and improve the other. And combining radar and ultrasonic sensors can provide data that neither can provide alone, no matter how good the individual sensors are.
But sensor fusion also works the other way, letting us peer deeply inside our cheap and imperfect sensors. Sensor fusion lets us dynamically perform characterizations and calibrations of imperfect sensors that greatly enhance the usability of their data.
Sensor fusion also adds robustness to the system: Failure of a big, expensive sensor can take down the entire system: Failure of one in an array of cheap sensors may trigger only a maintenance notification, with the system continuing to operate will within specifications.
I suspect the GM CEO is very hardwired when it comes to hardware, and is poorly informed about the power of software sensor fusion. Musk views his cars as software platforms, so he likely has a better grasp of the whole problem.
I've done lots of sensor development and sensor fusion for industrial and scientific systems. I haven't done a stand-alone single sensor (e.g., a lidar) in over a decade: My recent work is a diverse and redundant set of sensors, where the actual instrument is created via software. And I'm seeing ever improving performance for smaller increases in cost, a steady improvement in value per dollar.
I will take a few dozen cheap and imperfect sensors over a few "princess" sensors every time. Redundancy provides awesome benefits when combined via sensor fusion.
Neural nets, as they exist today, are nowhere near being able to "explain their thought processes". You can log what you fed to them, but you have no clue why they made the decision that they did, and there's no realistic way to trace it back. They're black boxes.
"If there was an antonym to 'Elon Musk', it would be 'Richard Branson'."
Autonomous vehicles could start with cargo in restricted road lanes ( barriers) . Musk / Tesla semi haulers are a good trial case. Soon can transition to remote roads etc... In parallel semi autonomous is helpful like anti brake Locks , cruise control, parking assistance etc
In ewe-lamb.cpp of course.
A human can do it with two eyeballs and a few mirrors... so what makes a machine different? I mean sure we'd prefer if the machines were safer than humans, but most of that is likely more to do with the software running on the humans than the lack of sensors... so the hard part is *not* sensors, it's interpreting the sensor data to safely move the vehicle.
You might not realize that you have general knowledge that (young animal from group) will tend to walk towards (older animals from group) but if you don't you're an idiot. And I'm sure you're not an idiot. You just aren't having a full appreciation for human intelligence, even when it is actively helping you make decisions while driving. Sure, sensors don't lose focus and people do, but really that is going to go a very small way. But will rule-based algorithms every be tuned to handle sheep, ducks, cows, moose, deer, foxes, wolves, etc etc etc? Very unlikely, because without a general understanding of how humans group and associate animals and a method to do that in a cpu without programming every shape and size, it will be a long if not impossible road to follow.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
These already exist, they are called free space detection networks. They don't detect that there is a sheep, they detect that there is no safe place to go where the sheep exists. Basically, they detect the road, and everything else is a no-go area.
All you need is two cameras on a gimbal, an accelerometer, and two microphones mounted coaxial to the cameras.
It's obviously possible to have a level 5 car operate on just cameras. A human driver can do it with two (or sometimes one) non-ideally positioned "cameras" and s few small mirrors. Of course there is a difference between "possible" and "practical with any current cor likely near-future technology", as to duplicate this would need a computer that can make a lot of inferences.
People do not actually use stereo vision for driving, the distances are too great. And we can drive fine with one eye closed. We do really clever processing to extract a 3D model out of one camera.
But computers based stereo vision works much better. The cameras can be further apart, and can be pointed to sub degree accuracy. The tricky part is realizing that a point on one image corresponds to a point on the other, looking in particular for vertical edges. Tricky, but much easier than reconstructing a 3D scene from one camera. Once that is done, simple trig gives you the distances. This tech has been around since the 1980s.
I have never understood the need for Lidar or anything else. Radar could be a good backup, if it sees something close then panic. Ultrasonics do not work in the wind, so hard to see how they could work in a car, other than parking. They are cheap though, I bought one on ebay for $2 the other day.
The louder and more obnoxiously you say it, the truer it is, right? Never mind that they just said that no human can drive since none of us have the sensory capability of LIDAR.
>> GM Exec Says Elon Musk's Self-Driving Car Claims Are 'Full of Crap'
GM execs are not relevant any more in the car industry.
aaaaaaa
We don't have "stitching errors" in how we build up a model of the world around us from visual data
What? Yes, we do. We do all the time. They don't matter at a walking pace. They do matter at a car's pace.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Humans do it with just sight, why wouldn't a property programmed computer that has a much faster reaction time, far better on the fly computational ability & other sensors (instant access to GPS, Speed, throttle position, brake level, etc) be able to? The difficult part is a system that can properly process the visual inputs. While it is obvious that the on-board computers don't have the capacity of the human brain they are only trying to replicate a small part of its function (driving a vehicle, recognizing objects). It's never going to be perfect mind you, but if you pay attention to history expecting something to be perfect in its first iteration is idiotic. You create something that is better than what came before it, nothing more and nothing less.
Absolutely nothing has changed with GM. Same old cars, same old company. They'll need another bailout as soon as the economy takes another dive.
You know, your argument could be taken as an example of why people are such bad drivers, and why self driving cars can be safer. They don't get overconfident. You Know Sheep. You look at them and say, "Those sheep are all on the same side of the road, so they're not likely to run out into the middle. They're not a hazard." You keep on driving without slowing down or worrying about them. The computer doesn't have your overconfidence in your ability to predict what sheep will do. It says, "Large animals by the side of the road are a hazard." It slows down and is ready to stop the moment they do anything unexpected.
Most of the time, you'll be right. The sheep will just stand there and you'll drive past. You'll be right 99 times in a row. That gives you a false sense of security and makes you less prepared to respond when the 100th time, a sheep does something you didn't expect.
"I'm too busy to research this and form an educated opinion, but I do have time to tell everyone my uninformed opinion."
There’s a huge difference between Old GM and New GM; Old GM got saddled by the pension obligations. New GM isn’t responsible for the pensions of the old employees of Old GM.
GM? Remember GM? GM quit the business and started the "New GM". Initially they tried to assume NO liabilities of the Old GM. (POTUS Obama was okay with the New GM screwing their customers, BTW.) The consumers caused such an uproar that the New GM and the POTUS had to back down on the liability scam. GM killed off some of their car lines to "streamline" the corporation. They tried to sell Saturn cars but refused to help the new buyer by continuing to manufacture cars for 2 year until the New Saturn could handle their own manufacturing. The New GM shafted the Saturn workers and owners big time. 100's of lost families and lost jobs. The Saturn dealers also disappeared -- more lost jobs. GM stock tanked and stockholders were left holding an empty bag. Nada. Nothing for their ownership and loyalty to the Old GM. Frak GM and POTUS Obama!
Tesla must be quite confident they can make it work, as they are already selling cars for which they claim full self-driving software will become available some day.
FTFY.
It seems to me we have forgotten about Google's Chaeffuer program. If memory serves that business is now part of WAYMO.
How many MILLIONS of miles have these vehicles driven to date?
-Brett
The car can certainly be programmed to recognize larger and smaller animals, and therefore it's easy to insert a rule about similar animals of different size on different sides of the road. We don't have to have separate rules for large and small sheep, cows, ducks, capybaras, pangolins, etc. This doesn't require intelligence.
Besides, for varying X, what are the triggers that will send young Xs running to their mothers? I don't know myself. Doubtless you know what will spook lambs, and that makes you a safer driver in your part of the world.
"When you have eliminated the unacceptable, whatever is left, however improbable, must be the truthiness" - Holmes
To a camera, an animal is a set of certain blobs of color arranged in a certain way with respect to one another. The arrangement of the blobs will change depending on the orientation to the camera. Once you are finished making the rules for a sheep, this will never work for a fox since a fox is a different set of colors. How will you ever make a rule based on a camera image that means 'it is an animal in general' when all you have to go by are color ranges, sizes, and positions? A racoon will look very different in this respect than a sheep, so different rules will need to be made. You could make a general rule to slow down for anything between 1' x 1' and 3' x 3' and moving, but then you'll be stopping for every tumbleweed and shopping bag blowing in the wind. This isn't as easy as you make it seem.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
It's not simple (very few things in computer vision are), but it's an animal recognition feature, not a behavior prediction feature.
"When you have eliminated the unacceptable, whatever is left, however improbable, must be the truthiness" - Holmes
The GM expert is saying that the HARDWARE needs to be fault tolerant. For example, the steering actuators had better be three way redundant. This is a detail the all of the other self-driving car developers haven't gotten to, yet. It is true that developing the software is the hardest part and you can do that without a fault tolerant hardware platform. But when it comes time to deploy, your hardware is fault tolerant or your name is Takata...
It would appear that GM is the only player out there who has realized that. So they might be a lot further along at deploying self-driving vehicles than M. Musk is. Or at least they don't plan on decapitating anyone as part of their development plan.
An engineer who ran for Congress. http://herbrobinson.us