Boston Dynamics' Atlas Robot Executes Autonomous Automotive Parts Picking (techcrunch.com) 67
In a new video published today, Boston Dynamics' humanoid robot Atlas is shown moving engine parts between bins without any human assistance. TechCrunch reports: Boston Dynamics is quick to note that the actions are being performed autonomously, without "prescribed or teleoperated movements." [...] Boston Dynamics notes, "The robot is able to detect and react to changes in the environment (e.g., moving fixtures) and action failures (e.g., failure to insert the cover, tripping, environment collision) using a combination of vision, force, and proprioceptive sensors."
In addition to the autonomously executed tasks, the video showcases impressive adaptive -- and strong -- actuators, as the robot pivots at its waist. The action minimizes movements, saving precious seconds in the process.
In addition to the autonomously executed tasks, the video showcases impressive adaptive -- and strong -- actuators, as the robot pivots at its waist. The action minimizes movements, saving precious seconds in the process.
This isn't a bad thing... (Score:4, Interesting)
At the minimum, this might reduce RSI/carpal tunnel issues with people. Having a robot handle the monotony of handling a box of parts and be able to do something with them is a good thing. The video included is good, but not too far off from what a tape robot already does in a tape autochanger... grip the tape, move it from one slot to another, although moving stuff and turning it 90 degrees 100% optically is a definite advance.
I'm all for parts picking robots. The less stuff that tears peoples' muscles and tendons apart with repeated actions that are not requiring any real skill, the better.
Robot Lego Grand Challenge (TM) (Score:2)
Hereby find the Robot Lego Grand Challenge (TM)
Stipulations:
1) Robot has to be autonomous and not remote controlled
2) Robot has to build a Lego Death Start model from a 4 foot x 4 foot box of individual Lego pieces
3) Lego pieces are not sorted, pre-built and are in a random pile in the box
4) Robot can be given a list of parts, and an order to build them, and the location where each part goes
5) Entire process must be filmed from multiple angles
Win condition: A fully built Lego Death Star
LEGO was years ahead of its time (Score:3)
I mean, already focusing on building a death star?
Re: Robot Lego Grand Challenge (TM) (Score:4, Interesting)
I recon you could make this much more interesting by saying the model is unknown before hand, and that the robot has to read the LEGO manual to build it.
Re: (Score:2)
This implementation would leverage this prior art, in a slightly reversed manner. The big obstacle I see would be creating a robotic pick-and-place appendage with sufficient dexterity to manipulate lego pieces.
Level 1 of LEGO build by Autonomous Robot (Score:2)
Let's back up a bit and start with a graduated scale of autonomous robot built LEGO sets
Level 1 - Something simpler with 10 to 20 larger pieces in a more simplistic build ...
Level 2 - 20 to 50 larger pieces in a more difficult build
on up to the Death Star.
There are good avenues for making it easier or more difficult by
1. Using very different high contract brick colors
2. Using only larger bricks versus a mix of large and small bricks
3. Having simpler or more complicated builds such as builds which have very
Re: Robot Lego Grand Challenge (TM) (Score:2)
Why are you making the challenge harder for the robot than a human? Having assembled a Logo Death Star twice I can assure you that the parts are sorted out of the box and it is way harder the second time when they are not sorted.
Reality Sets In. (Score:3, Insightful)
This less stuff that tears peoples' muscles and tendons apart with repeated actions that are not requiring any real skill, the better.
There is a portion of the population that requires employment to survive and ultimately retire. Today that would be basically all of us who were not lucky enough to be born into immense wealth. If you wish to start permanently removing human jobs that don’t require any “real skill”, then you had better be ready to start paying real taxes to support the unemployable masses.
Are you? Are you ready to start paying? Because as much as some may hate doing jobs like that, starving to death is
Re: (Score:2)
There is a portion of the population that requires employment to survive
The modern concept of mass wage-based employment has been around for less than two centuries. There's no reason to believe it's the only way God intended us to live.
We've been automating labor for three centuries. Since then, living standards have soared twentyfold, and unemployment is near a record low. This is more of the same, with more goods produced with less labor, meaning rising per capita wealth.
Re: (Score:1)
We've been automating labor for three centuries. Since then, living standards have soared twentyfold,
For the people not killed by it
and unemployment is near a record low.
If you believe the published unemployment rates, which no one should ever do, because they are based on known-flawed metrics.
This is more of the same, with more goods produced with less labor, meaning rising per capita wealth.
I would care about that if the wealth were evenly distributed, instead of ever more unevenly. Perhaps you missed the memo about trickle down not being a thing?
ahh here come the moderators to save the day (Score:1)
Poor defenseless crony capitalism, also known as corporatism, also known as fascism, needs some dildo too cowardly to respond to a comment who has to hide behind abuse of moderation to be its protector and savior!
Re: (Score:2)
We really don't need to stick children back in the mines, or humans for that matter, especially as most mining is done via robot. Lose a mining robot in a mine? Insurance claim. Lose a group of workers? The PR may hit world news. We really can't go backwards, and start "creating" jobs by having laborers come in and toil over crops in the field when a combine already does the work. All what will happen is that when companies are forced to hire people like this, offshore companies that are not burdened
Re: (Score:2)
Not everyone is trainable to actually critically think. Without better social safety nets in place, these "mundane" jobs that get replaced won't leave enough jobs for these category of people.
It's pretty awesome we can develop these capabilities but more and more we will displace expensive Americans. Sure, in the long run things will be better but in the short run, you risk hurting a lot of people because we won't be helping these untrainable people.
I'm not saying we should stop progress and I'm not saying
...but what about the displaced workers? (Score:2)
At the minimum, this might reduce RSI/carpal tunnel issues with people. Having a robot handle the monotony of handling a box of parts and be able to do something with them is a good thing. The video included is good, but not too far off from what a tape robot already does in a tape autochanger... grip the tape, move it from one slot to another, although moving stuff and turning it 90 degrees 100% optically is a definite advance.
I'm all for parts picking robots. The less stuff that tears peoples' muscles and tendons apart with repeated actions that are not requiring any real skill, the better.
I completely agree that using robots for repetitive, physically taxing tasks—especially ones that can lead to long-term injuries like RSI and carpal tunnel—can have real benefits for workers' health. Reducing the need for people to do work that’s often painful and potentially harmful is a smart application of technology.
One thing I’d add, though, is that while robots can relieve these physical strains, there's a bigger picture to consider around displaced workers. Even when automatio
Re: (Score:2)
Hopefully it is fenced off (Score:3)
Regular robots have killed people semi regularly for many years in industrial environments. The last thing anyone needs it is getting confused and going outside its workspace to smash, tear apart or roll over someone walking by.
Re: (Score:2)
Many items can be deadly including cement mixers, wood chippers, vending machines, etc. In fact, there are on average 70-95 deaths each year from lawn mowers. https://www.newsweek.com/lawnm... [newsweek.com]
As for industrial robots, the latest study I found was from 1992-2017. During those 25 years, 41 deaths were caused by robots - less than two per year.
https://www.safetyandhealthmag... [safetyandh...gazine.com]
As a side note, the first study found almost 200 deaths per year are attributed to deer. Over four times as many as are killed by li
Re: (Score:2)
Please put a fence around the deer, thanks.
Re: Hopefully it is fenced off (Score:3)
Anyone whoâ(TM)s dealt with deer in real life knows you can build a fence as high as you like, and the deer will find a way over, through or around it.
Re: (Score:2)
Uh huh, it was a joke. Carry on.
Re: (Score:2)
jokes operate on the same mechanism as deer, no matter the low bar to getting it, some deer never do.
Re: (Score:2)
My former mother-in-law in Montana hates deer for this reason. She grows raspberry bushes and has gone to great lengths to keep the deer away. They will stand on their hind legs to reach over if they don't jump. I think she was able to get some electric fencing put in that had more luck but it was definitely an arms race with the deer.
Hehe, I remember her commenting "Wish the hunters would shoot them all." Quite hard not to laugh at that remark.
Re: Hopefully it is fenced off (Score:2)
Re: (Score:2)
Uh that is basically impossible. It has to have multiple misrecognitions in a specific sequence for that to happen. I would have to think it's within its workspace, it will have to constantly recognize a human as a target workpiece. I don't see that happening easily.
Re: (Score:2)
Uh huh, it's an AI and it's really smart and it has all these fail safes to make it impossible to kill someone.
First, it would have to misidentify a person. Then, uh, yeah that's about it. AI has never gotten anything wrong. Right?
Or we could treat it the way we treat other dangerous industrial robots already: put a fence around it with numerous warning signs for humans to stay away because they're really fucking dangerous and sometimes kill people.
Re: (Score:2)
Regular robots have killed people semi regularly for many years in industrial environments. The last thing anyone needs it is getting confused and going outside its workspace to smash, tear apart or roll over someone walking by.
(Engineer) ”Hell no. I ain’t gonna tell him the damn thing is IoT enabled by default.”
Re: (Score:2)
A simple solution would be to keep the power lead nice and short.
If it must have a battery then a separate, fail safe system that cuts power when it leaves the range of a low power radio transmitter would work.
Re: (Score:2)
Regular robots have killed people semi regularly for many years in industrial environments. The last thing anyone needs it is getting confused and going outside its workspace to smash, tear apart or roll over someone walking by.
While caution is essential, framing all robots as “killers” is a gross exaggeration that doesn't align with the actual data [safetynewsalert.com]. Forty-one robot-related fatalities over 25 years represents a rate of under 0.02 deaths per 100,000 workers annually—a low figure compared to fatality rates in industries without high automation. For context, U.S. data from 2022 shows construction and transportation each have rates over 10 deaths per 100,000 workers, and fields like agriculture and forestry see rates
Re: (Score:2)
I didn't say all robots are killers. I was quite clear in plain simple English that they "semi-regularly" have killed and seriously injured people even with all the safety protocols in place. How many more injuries and deaths would there be without those protections?
There is a reason for boxing off robots and smothering them in warning signs. It reduces deaths and injuries.
Re: Hopefully it is fenced off (Score:2)
The robot can very easily be programmed to stay within a given area, this is a very old robotics feature. The problem is the humans that move into an unsafe work area. That's how most industrial accidents happen: human error.
Re: (Score:2)
It's not a traditional robot. An AI is running it. As we all know, we don't know how they actually work at a detailed level such that we can guarantee certain results in various situations. That's why we see "AI hallucinations" and other weird results commonly from so-called AI.
If you're going to program it like a traditional robot with so many guard rails it can't do anything different than a traditional robot anyway then you might as well just use a traditionally programmed robot without the AI. That'
AI (Score:2)
As great as Boston Dynamics has been, most of their algorithms seem to have been developed using traditional control system theory.
I wonder now that AI models can pre train themselves in their own physics world, whether some upstart will eat Boston Dynamics lunch with a robor based on neural net reinforcement training models?
Re: (Score:1)
I wonder now that AI models can pre train themselves in their own physics world, whether some upstart will eat Boston Dynamics lunch with a robor based on neural net reinforcement training models?
(Boston Dynamics, most likely) ”Hey, are we supposed to be talking about our advanced military development here?”
Re: (Score:2)
Over the years Boston Dynamics has been known to show off seemingly advanced capabilities which in reality were remote controlled.
Seems to be a pattern for a lot of high tech companies in the last decade: claim magical autonomous ML/AI capabilities but hire secret Indian remote operators to make it actually work, more often than not.
See also: autonomous cars, supermarkets without checkouts, chain
Re: (Score:2)
AI == Abroad Indians
Re: (Score:2)
This one does say "Fully autonomous" so I'd guess it's not remote controlled. BD does have a history of robots like spot that work autonomously already
Not On The Front Lines (Score:2)
Boston Dynamics keeps talking about not putting weapons in these things hands and setting them loose on the battlefield.
Nothing says they can't and won't be building missiles, or parts enough so two guys can screw them together in 10 minutes.
--
In this bright future you can't forget your past. - Bob Marley
Re: (Score:2)
Boston Dynamics robodogs have already been used on small scale by Ukrainian Army as front line reconnaissance units.
Considering how Ukrainians basically invented modern FPVs, I would expect that if they are cheap enough, we'll soon see these weaponized.
Over or under powered? (Score:3)
Pretty amazing, but feels like it is underpowered but not enough CPU or sensors? It is jerky, when it makes a mistake it is something like jamming a part full speed at the wrong angle. Yes it helps that it can pivot its waist (though the way its head can turn around like an owl or possessed person is pretty disconcerting), but it also takes a long time halted, seemingly to confirm that nothing is slipping or happening from an unprecedented direction (maybe it saw someone outside the camera's field of view). If you saw a person carry around a big metal plate cocked back behind their body as if ready to stab it forward, as opposed to holding it down by their hip and possibly swinging like a pendulum, you would be concerned they both don't know how to move economically and also are ready to clobber someone. Would it be smoother and less mistakes if it moved more slowly?
I think this is the Atlas robot that was so amazing doing gymnastics and running on inclined surfaces so this impression is very unexpected, it's acting more like a jerky parts picker (but way slower than actual specialized robotic parts pickers).
I am guessing this could put Amazon warehouse workers out of a job if they were cheaper than 3 shifts of workers, though specialized picking robots that stick to warehouse shelves might be faster for the picking part. Also the robot was not actually selecting parts and bins. Impressive though, seems to be able to pass a minimum warehouse work test for one test case anyway.
Re: (Score:1)
Unlikely, because this hasn't demonstrated to have bypassed the problem that halted Amazon's autopicker project. Grip strength modulation. In amazon's case, they couldn't figure out how to get picker to modulate grip strength for changing types of objects that need to be picked in a typical amazon warehouse so it doesn't crush some of them and doesn't drop some of them at utterly unacceptable rates.
In this demo, this problem is sidestepped entirely by using only flat plastic panels as objects being picked.
Re: (Score:1)
How come you haven't responded to your Russian lies Luckyo?
Did you realize the fine is as big as every single other person in the world realizes?
Re: (Score:2)
I expect they will rapidly improve it, like they did with mobility.
Elon Musk is probably pissing blood right now.
Re: (Score:2)
Elon Musk is probably pissing blood right now.
Thanks, that gave me a warm fuzzy feeling
Re: (Score:2)
It did seem like an oddly long pause after insertion of the test item into the slot. What was going on then? One thing that impressed me was that it moved the left arm back as a counterbalance as it reached forward with the right one to grasp and extract the item. And then it pulled the item out part way before stopping to get a better grip. I think a human might have used both hands in concert to pull the item out and push it into the slot. Maybe the bot can't do that yet.
I'm wondering how much power this
Read Between The Lines (Score:3)
Everybody is picking at whether it's a good parts picker. A dedicated dumb robot is better for many tasks than Atlas. We already know that.
That's not the real application they expect.
This is a "see, if it can do THIS, it can also do THAT".
What do you imagine THAT is?
Re: (Score:2)
aircraft carrier manual operations
Re: (Score:2)
Oh wait, you didn't mean that THAT...
Looks like same problem as amazon (Score:5, Informative)
They fail to demonstrate here that they have bypassed the problem that Amazon ran into a few years ago. Grip strength modulation.
Essentially what amazon discovered with its massive multi-billion automated picker project is that when you have to pick normal things that are being shipped, modulating grip strength not to crush significant percentage of objects and not to drop significant percentage of objects to be a borderline impossible task. Sensors and calibration needed and the speed at which this calibration must happen was just too hard.
Boston Dynamics sidesteps this problem entirely by using only flat plastic panels.
Fired. (Score:2)
If a human moved so slowly they'd be fired on the spot. The only reason Atlas isn't getting fired is people saw the Terminator movies.
Re: (Score:3)
Re: (Score:2)
That's a good point, but that capital expenditure matters. They'll need to cost $50K each or less, including the initial programming cost.
Re: (Score:2)
They'll need to cost $50K each or less, including the initial programming cost.
The programming is an NRE sunk cost. It is irrelevant when calculating cost-effectiveness.
Re: (Score:2)
Re: (Score:2)
Funny thing about machines is that it doesn't usually matter how slow they are, if they get the job done. If you can make a slow machine, it is relatively simple engineering problem to speed it up. By simple, I mean that you don't need to be a genius to do it, like you need to be when you invent the first computer.
And it only needs to be about 1/4 of human speed to be as fast as a human who works for 8h/day for 5 days per week. And even less if we include vacations, sick days, training etc. And even less, i
Executes? (Score:2)
While execute is a perfectly cromulent use of the word in this situation, perhaps 'performs' might have been a better choice. After all, using the words 'execute' and 'robot' in the same sentence is sure to rile up a certain segment of the population.
"Ya see there! Them thar robots be executing! Mabel, git yer gun. It's startin'."
L;ots of demos over the years.. (Score:2)
-- So little shipped robots.
So I wonder how many more years until they have an actual mass marker peoduct from them in wide use?
My guess is somwhere between more than a decade and never.
parking tickets (Score:2)
My brain wanted to read autonomous automotive parking tickets. Which is a place I might expect to see this thing, watching cars, watching the clock, watching the meter, printing tickets. Thing could pay for itself in a week in some places.
Why the pivoting head? (Score:2)
Is it just to make it more humanish? Wouldn't it make more sense to just have cameras in all directions and have it see 360 degrees? Someone smarter than me, please explain. Robotics is not my thing.
not the robot (Score:2)
This is computer vision, not robotics. People have been trying to do this for 50 years.
Hockey stick (Score:2)
It is so cool that the robot looks like it has had some damage from a blunt weapon and there is hockey stick near the robot, standing like a warning sign, to make sure that the robot obeys.