|
rudatron posted:Driverless forklifts are definitely going to kill jobs sooner than driverless trucks, because you have much finer control of the environment. A fully automated logistics train is kind scary, and is probably what Amazon is trying to aim for here. Kinda hoping though that other companies step up their game, because if that future is coming, it'll be bad if Amazon is the only player there. These already exist: https://www.youtube.com/watch?v=8gy5tYVR-28
|
# ? Jan 17, 2016 07:10 |
|
|
# ? May 2, 2024 12:11 |
|
rudatron posted:Driverless forklifts are definitely going to kill jobs sooner than driverless trucks, because you have much finer control of the environment. A fully automated logistics train is kind scary, and is probably what Amazon is trying to aim for here. Kinda hoping though that other companies step up their game, because if that future is coming, it'll be bad if Amazon is the only player there. Yeah they are already in use. Not Kiva bots - actual forklifts with a lidar bolted on top. I got a tour of a giant new warehouse about a year ago and half of their trucks were automatic. They still use human operators because the pallets are occasionally hosed up. The foreman was sure that would never change The elevator system storing the pallets on shelves was fully automatic, the tracking system was RFID and everybody carried around a computer that updated their work orders. All in all pretty impressive - no heavy lifting, machines for everything, very systematic. Giant robots roaming free https://www.youtube.com/watch?v=lDyi9Yipt7c edit: vvv For the ones in that video maybe. The ones I saw had a lidar and ran over RFID markers embedded in the floor. Bates fucked around with this message at 08:03 on Jan 17, 2016 |
# ? Jan 17, 2016 07:54 |
|
Anosmoman posted:
There are probably dozens IR cameras mounted on the roof for position tracking.
|
# ? Jan 17, 2016 07:59 |
|
Necc0 posted:These already exist: Automated Storage/Retrieval systems are cool too. https://www.youtube.com/watch?v=_n3nQymCx7U
|
# ? Jan 17, 2016 08:01 |
|
Necc0 posted:These already exist:
|
# ? Jan 17, 2016 08:43 |
|
asdf32 posted:Re:driverless cars This comment is uninformed bullshit, The Google car uses radar as well as IR and visible light, it can see through snow and rain. Decades away, lol whatever champ.
|
# ? Jan 17, 2016 18:38 |
|
rudatron posted:that is a lot of jobs being threatened, and all the people being laid off won't necessarily be able to find new jobs. Fundamentally the maxim that >= 1.0 jobs must necessarily be created for every job displaced is loving stupid. We've reached a level of society that would have been considered post-scarcity by someone just a century ago, and we use less and less manpower to do it. People need money to live, so we've created a massive service sector, but the day has finally arrived where that too can be automated. Even "complex" jobs like driving are starting to look pretty iffy. The service sector is way up, the participation rate is way down, and yields are way up. The writing is absolutely on the wall.
|
# ? Jan 17, 2016 23:15 |
|
Sethex posted:This comment is uninformed bullshit, But still doesn't drive in rain or snow. A recent article: Google’s self-driving cars pull over in the rain, while Tesla’s can self-park posted:Google says the self-driving cars will start exploring more challenging environments: “We’re beginning to collect data in all sorts of rainy and snowy conditions as we work toward the goal of a self-driving car that will be able to drive come rain, hail, snow or shine!” http://home.bt.com/tech-gadgets/future-tech/googles-self-driving-cars-december-report-tesla-software-update-71-summon-11364033025413
|
# ? Jan 17, 2016 23:22 |
|
asdf32 posted:But still doesn't drive in rain or snow. Lol even the section you quoted said they are already driving in rain and snow. Just they can't handle it 100% yet. Also if you read the thread I posted more evidence they can drive in rain earlier.
|
# ? Jan 17, 2016 23:36 |
|
Trabisnikof posted:Lol even the section you quoted said they are already driving in rain and snow. Just they can't handle it 100% yet. There are a lot of things "they can't handle 100% yet" and they need to solve them all before the hype of self driving cars is actually realized. And guess which things they've left for last? The hard things. The biggest misunderstandings here are from people who don't actually understand how development works. The last 1-2% is orders of magnitude harder than the bulk of the functionality. Most products just ship when they're 95% done because it's good enough. That's not going to work here. We're talking about cutting edge AI/Sensor processing in one of the most complex, variable, uncontrolled and unforgiving applications possible (far harder than aviation). When they say they're just starting to collect rain and snow data it says a lot.
|
# ? Jan 17, 2016 23:54 |
|
asdf32 posted:There are a lot of things "they can't handle 100% yet" and they need to solve them all before the hype of self driving cars is actually realized. And guess which things they've left for last? The hard things. Mainly it shows it doesn't snow much in SV or Austin. As the articles point out, radar doesn't get impacted by rain in the same way. Kangaroos are more challenging than deer but Google wasn't being lazy for testing in a deer heavy environment first. Is your argument the AI challenge is too hard, the sensors won't be good enough or some unforeseen problem will prevent adoption within +10 years of Google's overly-aggressive timeline?
|
# ? Jan 18, 2016 00:35 |
|
Trabisnikof posted:Mainly it shows it doesn't snow much in SV or Austin. As the articles point out, radar doesn't get impacted by rain in the same way. I think that his point is that there really isn't an elegant design theory for self-driving cars. It's not like theoretical physics. It's all hard-coded for the particular task and getting the edge cases in that kind of stuff right is the hardest part. And people will demand perfection and will not tolerate any bugs in this sort of technology. All the edge cases will need to be solved for there to really be the self-driving car that allows you to watch a movie or work during your commute. silence_kit fucked around with this message at 00:55 on Jan 18, 2016 |
# ? Jan 18, 2016 00:51 |
|
silence_kit posted:I think that his point is that there really isn't an elegant design theory for self-driving cars. It's not like theoretical physics. It's all hard-coded for the particular task and getting the edge cases in that kind of stuff right is the hardest part. And people will demand perfection and will not tolerate any bugs in this sort of technology. All the edge cases need to be solved for there to really be the self-driving car that allows you to watch a movie or work during your commute. While you were right that there is no elegant solution, autonomous vehicles will not rely on every edge case being pre-programmed. There are edge cases and use conditions they will need to be improved to meet, but the scope of the task isn't about thinking up every possible situation first. quote:What if...[insert hypothetical situation that could be tricky]? I think Google's timeline is overly optimistic to say the least. That 11 yr old will get a driver's license, but people are implying we need new technology when we don't.
|
# ? Jan 18, 2016 00:57 |
|
Blue Star posted:That's not going to happen, though. Not anytime soon. It may happen eventually but not for many decades. I don't think you know what you think you know: http://www.wired.com/2015/10/googles-lame-demo-shows-us-far-robo-car-come/ Google intends the car to be out of the prototype stage in the next 4 years http://www.reuters.com/article/us-adidas-robots-idUSKCN0SE1RL20151020 http://www.cnbc.com/2015/05/06/self-driving-semitruck-gets-license-to-ride.html Sethex fucked around with this message at 01:13 on Jan 18, 2016 |
# ? Jan 18, 2016 01:07 |
|
asdf32 posted:There are a lot of things "they can't handle 100% yet" and they need to solve them all before the hype of self driving cars is actually realized. And guess which things they've left for last? The hard things. It needs to drive better than people, that is it. Insurance companies will price insurance based on how much it costs them to cover the liability of insuring an auto. If urbanites use it like a cheap Uber you will get a bunch of young people adopting it, then as the technology improves so too will the market penetration, that doesn't have to take very much time at all given how the smart phone is like a wallet for people today. But sure, keep clinging. Sethex fucked around with this message at 01:12 on Jan 18, 2016 |
# ? Jan 18, 2016 01:10 |
|
Sethex posted:It needs to drive better than people, that is it. Driving better than people in all possible scenarios and situations is actually pretty hard to do. This is what asdf32 is trying to say. And self-driving cars will not only need to be barely better than humans to be accepted, they are going to have to do way better than humans for people to trust and adopt the technology. Sethex posted:If urbanites use it like a cheap Uber you will get a bunch of young people adopting it, then as the technology improves so too will the market penetration, that doesn't have to take very much time at all given how the smart phone is like a wallet for people today. You can't just release a self-driving car in a half-finished state and rely on customer feedback to improve the product. It will need to work perfectly out of the box. It is not like cell phones. silence_kit fucked around with this message at 01:33 on Jan 18, 2016 |
# ? Jan 18, 2016 01:18 |
|
silence_kit posted:Driving better than people in all possible scenarios and situations is actually pretty hard to do. This is what asdf32 is trying to say. And self-driving cars will not only need to be barely better than humans to be accepted, they are going to have to do way better than humans for people to trust and adopt the technology. Its actually not that hard. People are really bad at driving. There are already examples where Google's car might have gotten a better outcome than the driver did when they took control. Plus, the regulatory bar isn't "better than any person ever always", but D&D never seems to care about NHTSA regulations on autocars. Google's wrong about the timeline. And maybe wrong about people wanting the bubble cars. They might even be wrong about how to get to autocar. But people are vastly overestimating the gulf between current capacity and the level required for adoption. silence_kit posted:You can't just release a self-driving car in a half-finished state and rely on customer feedback to improve the product. It will need to work perfectly out of the box. It is not like cell phones. Tell that to Tesla. Also, to the degree you're correct, that's why there are milage requirements in the existing autocar regulations.
|
# ? Jan 18, 2016 01:55 |
|
Sethex posted:It needs to drive better than people, that is it. And this is the second point of confusion: The Goal. Some google engineers drink their own kool-aide and think the tech is actually going to be completely ironed out but the second problem with the publicly stated timeline is that silicon valley (and you) think society is going to accept these things on reason alone. First, I actually do believe, probably like you, that these things will be safer than humans in most circumstances within maybe 10 years. But that's not going to be the real life criteria for people accepting this tech. This isn't grandma behind the wheel, it's corporate america and when a corporation runs over a child it plays out entirely differently. The standards are going to be higher. Being as safe as humans is not the goal. And it takes time to prove out. Even if a literally perfect car existed today it would be a more than a decade before it passed the safety hurdles for widespread adoption. And that's assuming the costs are reasonable.
|
# ? Jan 18, 2016 02:13 |
|
Trabisnikof posted:Its actually not that hard. People are really bad at driving. There are already examples where Google's car might have gotten a better outcome than the driver did when they took control. If you want to talk about the performance of the self-driving car now, currently it is really horrible and at its best it is at the level of a teenager going through driver's education, with the constant sudden braking and causing cars to rear end it. asdf32's point is that that functionality is the easier part of automated driving to implement and that it is the last bit of functionality which is the hardest part. Trabisnikof posted:Tell that to Tesla. Tesla's autopilot isn't a self-driving car. However, it is part of a trend of incrementally incorporating more automation in cars. The biggest economic benefit of self-driving cars is when the human is not required to be paying attention or even there, and at that stage all the edge cases will need to be solved. silence_kit fucked around with this message at 02:35 on Jan 18, 2016 |
# ? Jan 18, 2016 02:20 |
|
quote:If you want to talk about the performance of the self-driving car now, currently it is really horrible and at its best it is at the level of a teenager going through driver's education, with the constant sudden braking and causing cars to rear end it. asdf32's point is that that functionality is the easier part of automated driving to implement and that it is the last bit of functionality which is the hardest part. Well luckily you're describing a learning machine, one with a perfect memory and a fleet of cars to learn from. So yeah, if it is a teenage, 10 years till it is a truck driver sounds about right. silence_kit posted:Tesla's autopilot isn't a self-driving car. However, it is part of a trend of incrementally incorporating more automation in cars, which is very likely what is going to happen. Unfortunately, the biggest economic benefit of self-driving cars is when the human is not required to be paying attention or even there, and at that stage all the edge cases will need to be solved. It is a "self-driving car in a half-finished state" since it still requires user there. People still want it. I'm not saying people are smart for doing so, but the idea that only the perfect autocar will be sellable is silly.
|
# ? Jan 18, 2016 02:28 |
|
I would have had found greater skepticism over the rapid and widespread adoption of Uber than an auto personal transit system. As it stands an app where a ride late at night that costs a fraction of a taxi and less than Uber would obviously be a consumer's first choice on day one. Your argument rests on the assumpotion that the world is populated by elderly reactionaries.
|
# ? Jan 18, 2016 13:46 |
|
Tesla's current cars don't have the sensor hardware needed for full automation btw. They are claiming their next iteration will be 'full autonomy ready'.
|
# ? Jan 18, 2016 15:07 |
|
Paul MaudDib posted:Fundamentally the maxim that >= 1.0 jobs must necessarily be created for every job displaced is loving stupid. We've reached a level of society that would have been considered post-scarcity by someone just a century ago, and we use less and less manpower to do it. People need money to live, so we've created a massive service sector, but the day has finally arrived where that too can be automated. Even "complex" jobs like driving are starting to look pretty iffy.
|
# ? Jan 18, 2016 15:22 |
|
https://www.youtube.com/watch?v=6ZDBjD-u3ds Raising the minimum wage increases automation!
|
# ? Jan 18, 2016 15:30 |
|
rudatron posted:Look at the US political system right now, look at the party with a clear hold on both Congress and the Senate (which it will probably keep till 2020, maybe 2024), and tell me how supportive they will be of a growing proportion of population that is unable to find employment. More people are going to fall through the cracks, and no one is willing to help them. America's system is completely hosed, though I am an outsider. Like legal bribery plus legal gerrymandering. The progressives seem to be Bernie's cliche in the sane category an then a pretty vocal subculture of people who treat identity politics like they are important issues. For me Trump is the only true American candidate, he represents what most Americans passionately are. They want for themselves, they are actually uninformed idiots, so insecure that they are fine being hosed over as long as the blacks an whatevers are well below their social status. An it is for this reason I think America is going to poo poo the bed when adapting to this next phase of capitalism. China being a majority shareholder in its entire economy won't have to really seize any means of production because they already did. Sethex fucked around with this message at 17:09 on Jan 18, 2016 |
# ? Jan 18, 2016 16:16 |
|
I missed this:Trabisnikof posted:While you were right that there is no elegant solution, autonomous vehicles will not rely on every edge case being pre-programmed. There are edge cases and use conditions they will need to be improved to meet, but the scope of the task isn't about thinking up every possible situation first. I'm sure that the Google car is much more hard-coded than the Google press releases let on. Certainly, you shouldn't conflate human learning with the types of statistical analysis that are programmed into the Google car, like in your following quote. quote:Well luckily you're describing a learning machine, one with a perfect memory and a fleet of cars to learn from. So yeah, if it is a teenage, 10 years till it is a truck driver sounds about right.
|
# ? Jan 18, 2016 17:30 |
|
silence_kit posted:I missed this: I don't know where you get that idea. Hard coding situations is both a bad idea for developing a strong learning machine and also counter to all the evidence of how google is developing their project. Do you have a source for that claim? quote:The bottom line is that we need to be ready and able to deal with all of it -- even if we’ve never seen that exact situation before. We use the same basic principles to guide us in all situations. If we’re not sure what something is or how it will behave, we’ll slow down to give ourselves time to gather more information about the situation. If it’s a moving object, we’ll categorize it as a vehicle, cyclist or pedestrian so we can make reasonable predictions about its behavior -- and be prepared for odd behavior. When in doubt, our cars will stop. Over time, we’re getting better at modeling and responding smoothly to the unexpected, from cyclists riding the wrong way in a bike lane, to the new fad of lane-splitting electric skateboards. quote:Responding to things we've never seen before: teaching a self-driving car to handle every possible situation it could encounter on the road is not feasible, as there’s an infinite number of possibilities. Instead, our technology gives it fundamental capabilities to respond correctly to unexpected situations as they happen, like when we encounter a dog-powered skateboard. quote:One of the beautiful things about our self-driving software is that it can get practice without ever leaving the garage. Using our simulator software, we can take any of the individual scenarios we practiced on our test track (as well as ones we encounter on the road) and subtly change the variations to get even more practice. What if that pedestrian were taller, or shorter? If they darted out into traffic at different angles? We can very quickly get virtual practice on hundreds or thousands of variations on a single on-the-street scenario, adding to the 10,000 to 15,000 miles of real-world driving experience we gain every week.
|
# ? Jan 18, 2016 17:43 |
|
And? Yes some general purpose reasoning get work in a lot of situations but the real world can cook up a dizzying array of complex scenarios which not only have to be dealt with safely but also efficiently. The automated car can't wait 20 minutes to make a left turn onto a busy road in rush hour. That's completely unacceptable from a user point of view.
|
# ? Jan 18, 2016 18:29 |
|
asdf32 posted:And? And? You're right? Good thing no one is planning or talking about the strawman autocar you keep describing. There's no denying that the problem is complex, but the idea that google has to preprogram responses to real world behavior shows a misunderstanding of how learning machines operate. They even have a team dedicated to coming up with strange and unforeseen situations to test the system with, e.g. people leaping out of a porta-potty and onto the road. If your argument is that a learning machine will never be good enough, I'd like to actually see some evidence to back up a large claim like that.
|
# ? Jan 18, 2016 18:38 |
|
Trabisnikof posted:And? No it's not a misunderstanding because "learning machine" doesn't mean anything. At all. All the work in AI is to come up with the algorithms to process data intelligently. If the algorithm isn't good throwing more data at it and "learning" doesn't matter. There are learning chat bots which have been up and running for decades collecting data and because they completely lack the necessary sophistication, they're not actually approaching anything close to intelligence. The chat bots were one of many examples of AI tech which showed rapid progress before hitting a wall.
|
# ? Jan 18, 2016 19:11 |
|
The current hurdle is getting the car to distinguish leaves and squirrels vs a different type of object. The machine already works generally an smoothing it out is certainly not the hardest part. we have people in this thread claiming decades away before anything changes that just sounds like rejecting change on an unconscious level. Reminds me of the stubborn rejection of climate change in american discourse.
|
# ? Jan 18, 2016 19:15 |
|
Sethex posted:The current hurdle is getting the car to distinguish leaves and squirrels vs a different type of object. Please tell us more about the ills of The Great Satan.
|
# ? Jan 18, 2016 19:17 |
|
asdf32 posted:No it's not a misunderstanding because "learning machine" doesn't mean anything. At all. Yeah and the fact you're comparing the work Google does on learning machines now to 90s chat bots does in fact indicate you might not be up to speed on the advances that have been made in the last several decades. Thus far you've actually not really articulated a rational why this task isn't feasible for google's learning machine backed approach. Your own inability to conceputualize the solution isn't actually a proof the solution isn't feasible.
|
# ? Jan 18, 2016 19:18 |
rudatron posted:Look at the US political system right now, look at the party with a clear hold on both Congress and the Senate (which it will probably keep till 2020, maybe 2024), and tell me how supportive they will be of a growing proportion of population that is unable to find employment. More people are going to fall through the cracks, and no one is willing to help them. Well, to get extremely cynical for a moment, a lot of people view that as a feature and not a bug. There's a reason a lot of post-scarcity sci-fi features planets with very, very low populations (say, a few thousand people). One not-entirely-improbable scenario is that the very wealthiest just allow everyone who isn't them to perish, as surplus to requirements. edit: for a practical example of this occurring, look at the current situation in Flint, Michigan: http://forums.somethingawful.com/showthread.php?threadid=3760458 One way to look at this situation, through a marxist lens anyway, is that Capital decided Flint was surplus to requirements, so the citizens of Flint don't matter, so the capitalist system is going to neglect them until they perish (die, move away, etc.). It's at least arguable that Flint could become the new normal for everywhere with an average income under the poverty line. Hieronymous Alloy fucked around with this message at 19:24 on Jan 18, 2016 |
|
# ? Jan 18, 2016 19:19 |
|
Trabisnikof posted:Yeah and the fact you're comparing the work Google does on learning machines now to 90s chat bots does in fact indicate you might not be up to speed on the advances that have been made in the last several decades. Thus far you've actually not really articulated a rational why this task isn't feasible for google's learning machine backed approach. Your own inability to conceputualize the solution isn't actually a proof the solution isn't feasible. Stop using the phrase "learning machine" as if that means something. Google neural net and you can be playing with a "learning machine" in a JavaScript app in 5 minutes. Second it's not impossible but it's simply a matter of comparing where they are to where they need to go (far) and layering those data points over historical examples of similar development processes (it's not a straight line). When you do that, you realize it's decades away. And when you add in the social/regulatory/legal problems that come on top it's crystal clear.
|
# ? Jan 18, 2016 20:03 |
|
asdf32 posted:Stop using the phrase "learning machine" as if that means something. Google neural net and you can be playing with a "learning machine" in a JavaScript app in 5 minutes. You're getting mad that I'm using an industry standard term because it is vague? Guess what, a neutral net != an svm != an AI. I am not privy to the kind of learning machines being used by Google, so I use the generic term. quote:Second it's not impossible but it's simply a matter of comparing where they are to where they need to go (far) and layering those data points over historical examples of similar development processes (it's not a straight line). When you do that, you realize it's decades away. And when you add in the social/regulatory/legal problems that come on top it's crystal clear. But the problem is, you're just making these claims based on "common sense" with little to no backup. For example, you cite regulatory issues as delaying technology, except regulatory frameworks already exist. You claim they have too far to go, yet can't really cite which kinds of problems will be damning. Also, you're being unclear by what you mean when you say "it's decades away". Looking at vehicle lifetimes, it is obvious a fully autonomous vehicle fleet is decades away at best. However, level 3 autonomous vehicles will be on showroom floors within the decade (at least Audi, BMW, Daimler, Ford, GM, Google, Kia, Mercedes-Benz, Nissan, Renault, Tesla, and Toyota think so). Level 4, might be 10+ years away, but I haven't seen any technically backed evidence it would take 20+ years. Do you have any?
|
# ? Jan 18, 2016 20:12 |
|
asdf32 posted:The automated car can't wait 20 minutes to make a left turn onto a busy road in rush hour. That's completely unacceptable from a user point of view. I think this problem will be solved by a secure networking standard that is only for cars. This would come before fully autonomous vehicles.
|
# ? Jan 18, 2016 20:17 |
|
McDowell posted:I think this problem will be solved by a secure networking standard that is only for cars. This would come before fully autonomous vehicles. Right but I would basically categorize this as an "infrastructure solution" in that the environment outside the car is being modified to facilitate its automation. In this case other cars would be modified to broadcast information and most other cars have to be on board. "Infrastructure solutions" like networking or like dedicated lanes or signaling or signage or whatever make lots of sense but are different than what is currently being advertised. And it takes lots of time to overhaul infrastructure.
|
# ? Jan 18, 2016 21:02 |
|
Why exactly can't autocars turn left with onboard sensors again? Remember that unlike humans they'll have correct information on the speeds and distances involved. Or let me put it this way, what evidence would convince you the technology will be coming to market with X years? For me, it is when the large automakers all set targets. Those will slip, but less than a decade. So level 3 for sale in most fleets by 2025 is conservative to me. Trabisnikof fucked around with this message at 21:23 on Jan 18, 2016 |
# ? Jan 18, 2016 21:19 |
|
|
# ? May 2, 2024 12:11 |
|
Trabisnikof posted:Why exactly can't autocars turn left with onboard sensors again? Remember that unlike humans they'll have correct information on the speeds and distances involved. Because the reality is you have to be a dick and cut people off sometimes which requires significantly better interpretation and prediction of data with lower margins of error than just determining "all clear". And this is exactly an example of a performance enhancement that may be orders of magnitude more difficult than the more basic functionality. Being overly cautious is how the Google cars are doing what they're doing right now but it's ultimately unacceptable for widespread adoption.
|
# ? Jan 18, 2016 21:25 |