Boston Dynamics' new Atlas robot doing work.

[BrianHG]

Finally, something you know isn't being puppeteered  by a human or following a prerecorded script.



All we need next is a video of Atlas making more Atlas....

[nctnico]

I find this impressive. Having been around for quite some decades already, I recall the first attempts of making robots like these but they never managed to impress. Bonus points for the developer who had the brilliant idea that the head and torso can rotate freely.

Elon Musk just ordered hundreds of these to replace those pesky factory workers so they have time to vote for his favorite idiot promising more jobs. 

[BrianHG]

I found the effortless really deep squat down at 1:03 while picking up an item in front of Atlas, not below, truly an impressive balancing act.
It's the first time Ive seen an automaton perform such a feat.

[pcprogrammer]

I find this impressive. Having been around for quite some decades already, I recall the first attempts of making robots like these but they never managed to impress. Bonus points for the developer who had the brilliant idea that the head and torso can rotate freely.

Indeed impressive, but watching it rotate like that feels weird.

Have to say that a human would outperform doing this task, but boredom will strike for the human, something a robot does not care about.

I wonder how the robot was instructed to perform this task and how it decided to do it in a seemingly random way. A human worker would be told to fetch the items from the racks and place them in the other container, and go ahead with the job.

As a side note, this is something the "realistic humanoid robot" on this forum will never be able to do. 

[SiliconWizard]

That's impressive robotics, and otherwise nice marketing, but adapting the task and work environment for simpler and much cheaper robots is much more effective usually. So for work? Yeah. Sure. If you want to sell "leaders" the idea of directly replacing human workers without having to make any change to the work environment and kind of tasks. That's odd or very short-term.

There are areas where they can be a benefit though, like in risky areas with nasty chemicals, radiation, etc, when a human body is otherwise well adapted.
But this probably sounds much too "niche" to be good marketing.

As to the robots not caring about being bored. They may eventually get more rights than mere humans in the great industry 5.0 world.

[ebastler]

Finally, something you know isn't being puppeteered  by a human

How do we know that? I would definitely assume that the robot autonomously coordinates its movements to bend forward, squat down etc. But is it doing all the image recognition, action and motion planning etc. on its own, or might there be a human somewhere across the room with a camera feed and some custom controllers?

[AndyC_772]

The tech looks very impressive, but it's still very apparent that the actual job it's doing is an absurdly simple one, like the kind of thing the player might have to do in a 1st person puzzle game. It's at that level of complexity - stand near collectable item, 'take', stand near target area, 'drop'.

I'm sure I've done something very similar in a VR game, and it still feels a million miles away from the kinds of work that humans perform in reality.

Does it have anywhere near the detailed object recognition and fine motor skills to, say, assemble a PCB into a housing and plug in the wiring loom? Or pack a shopping bag? Feed the cat?

[mikeselectricstuff]

I find this impressive. Having been around for quite some decades already, I recall the first attempts of making robots like these but they never managed to impress. Bonus points for the developer who had the brilliant idea that the head and torso can rotate freely.

Elon Musk just ordered hundreds of these to replace those pesky factory workers so they have time to vote for his favorite idiot promising more jobs. 

Why would he when he has Optimus - I'm waiting for the video of an Atlas vs.Optimus cage-fight

[squadchannel]

“Fully Autonomous” in the lower right corner is not enough evidence, but it is great.
In terms of managing the factory, it cannot be completely standalone, so it must be communicating in some way.
but if Atlas itself is doing all the image processing and motion control, this is great.

I would have liked to have seen it demonstrated with a jammer. I think can prove that it is standalone.

It is truly amazing to see more and more robots like this.

[guenthert]

The tech looks very impressive, but it's still very apparent that the actual job it's doing is an absurdly simple one, like the kind of thing the player might have to do in a 1st person puzzle game. It's at that level of complexity - stand near collectable item, 'take', stand near target area, 'drop'.

I'm sure I've done something very similar in a VR game, and it still feels a million miles away from the kinds of work that humans perform in reality.
[..]

   Does it?  In the eighties school made me visit a factory where they made pots for plants from clay.  I still starkly remember one work place where a woman took a workpiece from one shelf on the left, placed it under the press, pressed two buttons (you know why), turned and put the now pot-shaped workpiece on a shelf to her right.  Again and again and again.  I was so shocked, I couldn't move for several minutes.  I henceforth put decidedly more effort into my education (arguably objective of the school trip achieved).   
   Perhaps that has been fully automatized in the past forty years, as robots became cheaper than even unskilled labor and that poor soul has been freed of her servitude, but I'm sure there are still millions who toil like that five or (soon again in Greece) six days a week.

[xrunner]

   Does it?  In the eighties school made me visit a factory where they made pots for plants from clay.  I still starkly remember one work place where a woman took a workpiece from one shelf on the left, placed it under the press, pressed two buttons (you know why), turned and put the now pot-shaped workpiece on a shelf to her right.  Again and again and again.  I was so shocked, I couldn't move for several minutes.  I henceforth put decidedly more effort into my education (arguably objective of the school trip achieved).   
   Perhaps that has been fully automatized in the past forty years, as robots became cheaper than even unskilled labor and that poor soul has been freed of her servitude, but I'm sure there are still millions who toil like that five or (soon again in Greece) six days a week.

Look at the work people do in countries like Pakistan - it's all easily found on YT. If anyone wants to be shocked at what people do all day for a meager wage watch a few of them. Untold numbers of these jobs could be taken over with robots like that.

And of course that leads to the other question what happens to all the jobs like that when these robots take them over - and they will. A worry for another day I suppose.

[ebastler]

Look at the work people do in countries like Pakistan - it's all easily found on YT. If anyone wants to be shocked at what people do all day for a meager wage watch a few of them. Untold numbers of these jobs could be taken over with robots like that.

But for tasks of the limited complexity shown in the video, there is no reason at all to use humanoid robots with all their complexity, stability challenges etc. One can easily envision a small family of generic, much simpler robots which can be programmed to handle many tasks of this type. The following should cover most of it:

- Just a 6 DOF robotic arm, stationary
- Robotic arm on wheel base
- Pair of robotic arms, stationary
- Pair of arms on wheel base

[xrunner]

Robots don't get sick and don't need to be paid and can work in areas where the need to have environmental conditions acceptable for humans can be removed (i.e. air conditioning and removal of toxic air). They will last as long as you have spare parts and won't revolt or complain or steal or do poor quality work do to exhaustion. So yes I do believe it will happen in poor countries because there are the rich there too.

To paraphrase a quote from the Oracle in The Matrix: "What to people with a lot of money want? More money".

This can be argued all day long but let's see what happens. I believe I will be correct but time will tell.

[tom66]

Robots don't get sick and don't need to be paid and can work in areas where the need to have environmental conditions acceptable for humans can be removed (i.e. air conditioning and removal of toxic air). They will last as long as you have spare parts and won't revolt or complain or steal or do poor quality work do to exhaustion. So yes I do believe it will happen in poor countries because there are the rich there too.

To paraphrase a quote from the Oracle in The Matrix: "What to people with a lot of money want? More money".

This can be argued all day long but let's see what happens. I believe I will be correct but time will tell.

The problem is a robot like this will cost, say $100,000, right now.  If it has a 10 year useful life then the minimum wage worker in China is safe for now.    If however the prices fall (as most technology costs do), then it will be a problem.

To a lesser extent this is already being seen by chatbots replacing low level tech support.  These bots may eventually replace voice calling, which will devastate the services economies in India, Philippines, Indonesia etc. 

We are going to have to get rid of the idea of full employment being an achievable goal.  That creates major headaches for capitalism which relies on consumption to survive.

[tszaboo]

The problem is a robot like this will cost, say $100,000, right now.  If it has a 10 year useful life then the minimum wage worker in China is safe for now.    If however the prices fall (as most technology costs do), then it will be a problem.

To a lesser extent this is already being seen by chatbots replacing low level tech support.  These bots may eventually replace voice calling, which will devastate the services economies in India, Philippines, Indonesia etc. 

We are going to have to get rid of the idea of full employment being an achievable goal.  That creates major headaches for capitalism which relies on consumption to survive.

100K is quite cheap to replace a human in a production environment. At some point my home country was trying to get manufacturers to open factories there. They always told the amount invested and the number of jobs created, and I just divided them. In my estimation it was always around half a million EUR or USD to create one such job. That of course includes the cost of the building and the production line, but still, it's money spent vs job created.

That being said, in accordance with EU law, you cannot have moving parts such as this shared with other workers in the same area. So for example right now I don't replace a worker with a Kuka robot arm, because all the paperwork, compliance, rules and everything I would need to go through. Plus I'm not a production engineer. But for a dedicated factory floor with said robot arms and moving parts, the paperwork is actually less.

[xrunner]

The problem is a robot like this will cost, say $100,000, right now.  If it has a 10 year useful life then the minimum wage worker in China is safe for now.    If however the prices fall (as most technology costs do), then it will be a problem.

Yes agree. We don't know many things at the moment, we'll just have to see how it plays out. All I'm saying is if it does play out that inexpensive robotics (it doesn't have to be robots that look like people either) take over many menial jobs, the world is going to have a grave problem to solve with employing all the people left behind.

[SteveThackery]

I agree - the "work" in this case is contrived and trivially simple. But the video is intended to be a showcase for the robot behaving in a human-like way, with its highly sophisticated movements, balance, and so on. The "work" also seems to showcase some fairly basic object recognition, along with a degree of planning which is required for the robot to move to the right place, in the right position, to deposit the object.

In reality you would never use a humanoid robot for a basic pick-and-place role. It's just a demo, and a very impressive one.

[themadhippy]

wonder how many miles of fishing line and how many thumb tacks are in that thing

[CatalinaWOW]

Obviously a humanoid robot is the wrong answer for an assembly line that will be doing the same task for months or years.  But there is opportunity in short run production.  Also fabulously more difficulty.  Humans are probably safe in that arena for decades.

[tom66]

What I want to know is how easy is this thing to train.  Did they say "move the engine covers from here to there", or did they say "move item that looks like (deep learned engine cover) from location 2A to location 2B, iterate until 8 covers have moved, incrementing the counts as we go, 2A is marked like this, 2B is marked like that"?

I think this distinction is crucial because if you ask a human to move the engine covers from one area to another they will "solve" the rest of the problem themselves.  Even the dumbest person can do that without further instruction.  If you have to supervise and carefully instruct this thing, it'll take more resources than it saves, even if the motions to implement the actions themselves are fully autonomous. 

[nctnico]

I think this distinction is crucial because if you ask a human to move the engine covers from one area to another they will "solve" the rest of the problem themselves.  Even the dumbest person can do that without further instruction.

Don't be so sure about that. In order to keep things organised, you'll need to give people very detailed instructions and monitor whether they follow those instructions. You'd be surprised how many people can't do the simplest jobs and / or have a different idea compared to your own idea.

[globoy]

Probably this version of Atlas won't enter the work force in large numbers...

But I see it like the very early personal computers, and early personal computer software.  These machines weren't ever going to be adopted in large numbers, but they were a necessary stepping stone to the point where the personal computer was ubiquitous and deeply embedded in society a number of years later.

Think about the rate of change in bipedal robots over the last 10-15 years, from Honda's Asimov series to this.  Now think about what's going to happen over the next 10-15 years given the rate of change in the AI world and the amount of investment currently into the marriage of AI and robotics.  One might argue that the bipedal form isn't always the best for functional robots but it certainly is one that captures the imagination of humanity and is getting significant effort.  I'd bet that some of us will have robot caretakers in our dotage.

[HuronKing]

" We designed Atlas to be capable, intelligent, and robust, so even when a pesky pickle moves a dolly, the robot can autonomously recognize its new position and complete its task."

[ebastler]

That motion at 0:51 makes me slightly nervous... You wouldn't want to annoy that thing to the point where it gets impatient with you. 

[SteveThackery]

I think this distinction is crucial because if you ask a human to move the engine covers from one area to another they will "solve" the rest of the problem themselves.  Even the dumbest person can do that without further instruction.  If you have to supervise and carefully instruct this thing, it'll take more resources than it saves, even if the motions to implement the actions themselves are fully autonomous.

In a way, that's a slightly separate issue. The USP of Boston Dynamic's work is in the name: dynamics. They've made a bipedal, humanoid robot that can walk over rough - even loose - terrain, dance, jump, run, get up from the floor, stay balanced when someone tries to push it over..... it is magnificent, world-leading work and they have transformed the world of robotics.

The stuff you describe, which is really about machine intelligence, is something BD are working on, but don't seem to be the class leaders as far as I can make out. I think many, many commercial and academic organisations are working hard on it.  I think we would agree that if you wanted a robot to move items from one rack to another, you'd be nuts to invest in something as versatile as Atlas.  A box on wheels with one arm would be far more practical. And as you say, the "intelligence" of the machine - which would apply equally to Atlas and the one-armed box - is important but not obvious from the demonstration video.