Robots still suck. It's all they can do to stand up – never mind rise up Just after lunch on a sweltering summer day in Brisbane, Australia, a dozen scientists and engineers gathered to watch a dog named Bingo stand up and trot gingerly towards a man-made tunnel. At the entrance, Bingo stopped to 'think' for a minute or so before turning its body to walk inside. Bingo was then joined by four …
The Elon is presumably well aware of the capabilities and limitations of robots. His attempt to manufacture the Tesla Model 3 largely with robots wasn't a total failure. But it was anything but an overwhelming success see https://www.theverge.com/2018/4/13/17234296/tesla-model-3-robots-production-hell-elon-musk The factory did end up using a lot of robots we're told. But apparently he ended up with many more human workers than were in the initial plans.
Most car plants are by robots BUT these are not some sort of semi thinking robot, they just repeat a series of movements they are told to do. A few have sensors that allow a minor adjustment or two to the movements. Looking at videos of say the mini plant in the UK gives some idea why it is NOT the case that the cost of human workers in modern production sites has any real relevance. What are problems start with cost of land (far far far far far too high in the UK and only going up), cost of energy (ridiculous in the UK and going up) and the cost of capital (again far too high - largely due to so much being sucked up for the purchase of non productive housing stock - which is also responsible for pushing up the cost of land, denuding the UK of holiday places and pushing the UK price of holidays through the roof).
To regenerate industry in the UK is a very very simple thing. The government needs to tax the difference between the price paid for a home and the price it is sold at by 150%. It needs to stop giving planning permission to convert factory sites and holiday camps to housing. (Cowley in Oxford was a huge car manufacturing plant, its now a housing estate... more profit in the short term)
On the road, humans are given stop lights, lines down the middle of the road, signs every now and again to remind them of low bridges and the lollypop man.
Asking robots to perform with infra designed for humans is as pointless as expecting humans to read the speed limit off a barcode.
Its not, following lines is something most robots can do - even those you can buy for kids... thats not difficult. Working out which line to follow when there are multiple lines is a problem. If computers are supposed to be able to recognise obstacles in the road then they should be able to recognise signs - there are fewer types to understand.
AI finds textual answers to textual questions, where both questions and answers are clarified by their textual contexts (they are annotated). In other words, AI compares textual contexts. Comparing this kind of contexts AI receives textual instructions, explaining what to do.
Thus robot can chose one line among all and is “able to recognise obstacles in the road”, based on text.
On the road, humans are given stop lights, lines down the middle of the road, signs every now and again to remind them of low bridges and the lollypop man.
And even still, humans ignore things like large lit signs warning them that the truck they are driving won't fit under the bridge.
I've been telling people if there was actual intelligence (as opposed to pre-programmed pattern recognition) in "self-driving" cars, they wouldn't rear-end firetrucks. But then an acquaintance actually did exactly that -- rear-ended a firetruck, almost exactly like a few Teslas have. I'm not sure what to make of that.
"Automated guided vehicles, such as driverless forklifts, are common in modern warehouses but need dedicated infrastructure to operate properly. One wrong QR code, or a shelf that's a few millimetres outside a set tolerance, and they struggle."
A guy I once worked with had managed an automated warehouse using high lift robot forklifts working racks some 25 feet high (admittedly somel decades back). One day the bar code database died and everything stopped dead. They had to bring in mountaineers to abseil over the racks recording all the bar codes afresh.
Well, at least they're good for vacuuming - maybe.
In any case, it seems that Real Humans are still a good ways off.
Maybe that's a good thing.
>> Sometimes even a few blades of tall grass can stop a robot in its tracks.
Sounds just like a horse to me... Sunlight? Butterfly? Grass? Gate? yup - all will spook a horse triggering any or all of freeze, flight or fight depending on the euqine involved..
I reckon these so called robotic canines are actually equines in disguise.
Perhaps they are programmed by young riders in jodhpurs... /me starts humming Velvet Green..
nurse! nurse! get the medication!
I've never heard of a gilly suit. But a google[1] search returned this: Ghillie suit, a type of camouflage clothing. Still, thanks for the reference. It is always good to increase my knowledge. :-)
[1] Well, duckduckgo. I never use google search.
Robots are perfect workers:
- dumb, they'll never question their manager
- only need food / mains. They won't feel jealous about colleague new iPhone.
- employer can legally pay nothing
- robot will unlikely have an affair and cause a scandal
- robot will not join a union
There are some bad things, like robot cannot use NHS and you have to hire a private robot doctor if something goes wrong, but that's pretty much it.
- dumb, they'll never question their manager
Dumb unquestioning adherence to instruction is the best way to rebel
- only need food / mains. They won't feel jealous about colleague new iPhone.
Ok, I'll give you that one ;)
- employer can legally pay nothing
Except support, enhanced support, feature requests, cloud resilience, etc, etc
- robot will unlikely have an affair and cause a scandal
Not even with the photocopier?
- robot will not join a union
Have you never noticed how when one thing breaks, other machines stop working in sympathy? They already have a union!
- robot will unlikely have an affair and cause a scandalNot even with the photocopier?
- robots will not join a union
Lester: Why are we doing this?
Rimmer: What are you talking about? This is vital, absoutely vital work!
Lester: Come on, we all know that the only reason we're stuck doing this and not the maintenance droids, is because they've got a better union than us.
Red Dwarf already called it! Not only will Robots join a union, they'll have a better union then their meatsack colleagues...
"In Germany the employees are robots, repeatedly doing stuff and never questioning or challenging,..."
Having spent my working life in Germany, I do not recognize this description. Can you provide some sources?
A fairly interesting review of the VW factory: https://youtu.be/WlSxy_5GGh0
(With which I am not affiliated)
Dunno about that - after pub o'clock it definitely gets a bit trickier to achieve.
Then again, getting back to robots, this is really quite mesmerising: https://www.youtube.com/watch?v=ssZ_8cqfBlE&t=5s
The final line was exactly what I was thinking while watching the video, the entire grid is one robot.
The complexity is a hurdle to overcome in designing it from the start but that complexity is why it works.
Plus, as clever and complex in appearance it may serm, the overall system is relatively dumb in AI terms.
Especially now as summer is here. In cooler times my clothes act as lubrication and allow me to brush past a door frame when navigational error lead to a slight collision. I've notice that bare flesh drags and ruffles up to make a slight collision into a face slammed against the wall incident. I think maybe that's why flock wallpaper got popular.
Arguably, the problem is not that robots are dumb, the problem is that we don't know how to teach them or to give them appropriate sensory apparatus to learn. One huge advantage for the robots will be that they can exchange information much more rapidly than we can, so whatever they do learn can be disseminated across the entire population or a significant subset thereof. We can therefore assume exponential learning capabilities.
That should be fun.
It's the old problem. If we could sufficiently define what we want and how it works then we might be ok - but we can't.
I can't remember which AI pioneer it was who, when challenged that AI could never be intelligent, replied "You define intelligence and I'll build you a machine that does that".
Our problem is we are starting from scratch. We should really understand that animals have been doing this for 600,000,000 years or so and have systems we can probe for clues on how to solve the problems. Or maybe we are like those robots and just a bit to dumb to work out how evolution has managed it.
Tell that to the ATLAS. The threat is that our world is about to be over run with robots controlled by an over arching, all knowing, all seeing AI. One that has you down to our DNA, and retinal scan, and knows exactly what scares you (mostly the AI it'self rn). Your world is being artificially made and fed back at you in a virtual feedback loop, and its only getting worse from here in a hurry. They have achieved WONDERS in 20 yrs. lol
What kind of artificial WONDERS await us in the next 20?
$75,000 for a robot the size of a dog?. That will buy a hell of a lot more of the canine variety, even at lifetime costs. Given that the lifetime of mechanical objects, of even low complexity, the canine version will easily outlive it's robot equivalent. For the larger, load carrying versions, The British Army in the far East, during ww2, employed mules, at a far lesser cost, for a far greater effect. Going back to civvy times, the old milk delivery, by horse and cart, utilised the intelligence of the horse to follow a route, stopping where required, without apparent input from the human controller. It might be interesting work, and might tick the "Gooodness Me!", box, but it's hardly cost effective.
Some friends of mine had a pony and trap and frequently woke up in their barn after a good session down the pub a few miles away. The pony would even wait patiently for them to get back in the trap when they'd fallen out asleep, which normally but not always woke them up. I should warn anyone wishing to try this that trying to get onto a trap when pissed is incredibly difficult as the suspension on the thing means it all moves when you apply weight and tugs to try and get aboard. I would hazard a guess that for this to succeed you have to come from a horsey family and have been climbing into traps since you were able to walk.
I find it hugely amusing that people are now saying what has been obvious for a long time that using computer algorithms or 'learning' to navigate is difficult for robots, yet at the same time people (Musk, BBC and others) are all claiming we will have self driving cars etc. n a few years. Its bollocks, the issues are only just becoming clear to those management types that make these claims and its not going to be possible for some longish tie to come.
Last Tesla crash seems to point to Teslas software not predicting the path of stuff around it so being unable to predict when it needs to adjust its trajectory. A vehicle pulled out in front of a Tesla (the driver shouldnt have) and the Tesla didnt respond at all to the problem until the Ford was actually in its lane - a fraction of a second before collision occurred. Heavens that is a mistake in the design of the software so fundamental I wonder if the guy coming up with it had ever sat in more than a rear facing child seat before?
The article states "Automated guided vehicles, such as driverless forklifts, are common in modern warehouses but need dedicated infrastructure to operate properly."
That used to be the case for Automatic Guided Vehicles (AGVs), which followed lines painted on the floor, wires embedded in the floor, or bar codes, etc. and constantly communicate with a server.
However the modern units transporting pallets, etc. in warehouses are referred to as Autonomous Mobile Robots (AMRs). They initially survey the warehouse with their LIDARs, build up a map which is tidied up by humans, and then use that map to navigate, again using LIDAR, and only communicate with the server to get the start and end points of their trip.
There are quite a few companies supplying hardware and software for this. For example, www.bluebotics.com provide software and navigation hardware which is used by a number of AMR suppliers.
May all your pallets be transported safely, especially if they're stacked with -->
Worth a watch on YouTube. He talks about AI safety a lot. And the Specification Gaming episode was quite funny. A list is here
https://docs.google.com/spreadsheets/u/0/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml
I specifically like
"Road Runner Agent kills itself at the end of level 1 to avoid losing in level 2 Saunders et al, 2017 Trial without Error: Towards Safe RL with Human Intervention"
And
"Block moving A robotic arm trained using hindsight experience replay to slide a block to a target position on a table achieves the goal by moving the table itself. Chopra, 2018 GitHub issue for OpenAI gym environment FetchPush-v0"
So will it get to a point when, like Skynet, the AI decides "To make it appear I've done my daily task correctly, I'll just kill the human instead"
"Cobots are being implemented around workers at manufacturing workstations to inspect for faults in the product as it is being built"
- <deep intake of breath through clenched teeth>
- "You missed a bit"
- "You don't want to do it like that"
I reckon it would be about half a day before the cobot got picked up and thrown out of the window.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
