Slashdot Mirror
Engineers Create a Robot That Can 'Imagine' Itself (eurekalert.org)
Columbia Engineering researchers have made a major advance in robotics by creating a robot that learns what it is, from scratch, with zero prior knowledge of physics, geometry, or motor dynamics. Initially the robot does not know if it is a spider, a snake, an arm -- it has no clue what its shape is. After a brief period of "babbling," and within about a day of intensive computing, their robot creates a self-simulation. The robot can then use that self-simulator internally to contemplate and adapt to different situations, handling new tasks as well as detecting and repairing damage in its own body. From a report: The work is published today in Science Robotics. To date, robots have operated by having a human explicitly model the robot. "But if we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators, then it's essential that they learn to simulate themselves," says Hod Lipson, professor of mechanical engineering, and director of the Creative Machines lab, where the research was done.
For the study, Lipson and his PhD student Robert Kwiatkowski used a four-degree-of-freedom articulated robotic arm. Initially, the robot moved randomly and collected approximately one thousand trajectories, each comprising one hundred points. The robot then used deep learning, a modern machine learning technique, to create a self-model. The first self-models were quite inaccurate, and the robot did not know what it was, or how its joints were connected. But after less than 35 hours of training, the self-model became consistent with the physical robot to within about four centimeters. The self-model performed a pick-and-place task in a closed loop system that enabled the robot to recalibrate its original position between each step along the trajectory based entirely on the internal self-model. With the closed loop control, the robot was able to grasp objects at specific locations on the ground and deposit them into a receptacle with 100 percent success.
For the study, Lipson and his PhD student Robert Kwiatkowski used a four-degree-of-freedom articulated robotic arm. Initially, the robot moved randomly and collected approximately one thousand trajectories, each comprising one hundred points. The robot then used deep learning, a modern machine learning technique, to create a self-model. The first self-models were quite inaccurate, and the robot did not know what it was, or how its joints were connected. But after less than 35 hours of training, the self-model became consistent with the physical robot to within about four centimeters. The self-model performed a pick-and-place task in a closed loop system that enabled the robot to recalibrate its original position between each step along the trajectory based entirely on the internal self-model. With the closed loop control, the robot was able to grasp objects at specific locations on the ground and deposit them into a receptacle with 100 percent success.
When you put it in exactly those terms, I'm forced to wonder, WHY THE FUCK DO WE WANT THAT?!?!
"Prediction: within 10 years, Windows will be a Linux distribution." Me, 7-6-2016
Most of the time when I read a machine learning or AI story it seems fairly benign and reasonable. Other times I feel a pit form in the bottom of my stomach and wonder how close to the tipping point we our to creating our robot overlords.
----- obSig
Robots imagine it's self? Somebody has a vivid imagination..
I'm guessing it's not the robots...
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
Does it identify as human and are we supposed to not just accept that but celebrate it and bestow upon it the pronoun of its choice assuming that it further chooses to identify itself somewhere along the gender spectrum?
Sheesh .. I thought everyone knew that you shouldn't anthropomorphize machines .. they don't like it when you do.
I am Slashdot. Are you Slashdot as well?
This does not know what it is. It has just figured out parameters of the neural network that make it act according to a (human) model of its physics. That's not anything like self awareness. It's all numbers, plain math.
You mean a machine performed self-calibration? Welcome to 2019 style "engineering".
Sonny(iRobot)? The Robot(Lost in Space)? hal 9000(Space Odyssey )? Maximilian(The Black Hole)? âR2-D2/C3PO (Star Wars)?
Damn, I haven't even come close to touching all of the different robots that have previously been named.
Anonymous comments are as pathetic as the anonymous "sources" that contaminate gutless journalism from the New York Time
I think this was your first mistake.
Ascalante: Your bride is over 3,000 years old.
Kull: She told me she was 19!
Skynet was supposed to become self-aware August 29th, 1997.
Have gnu, will travel.
It will be just fine.
Because, surely, nothing will ever go wrong with self-aware, intelligent, self-repairing and self-replicating robots.
A computer ("robot") can no more imagine itself, let alone understand the contents in it than can a thermostat or a piece of paper with spreadsheet printed on it.
We have anthropomorphized the computers. They are simply not capable of Experiential Knowledge.
This is a waste of our energy and ability to deal with how computers will actually cost people their jobs and livelihoods.
So, we end up with a robot which has developed some primitive concept of what it is shaped like, how it can move, and its orientation in space.
OK, now what?
It doesn't know how strong it is, it doesn't know what people are, it doesn't know you can't just pick up people and stuff them into holes.
Now it's just this .... thing which can move, but otherwise has no parameters or constraints on what it is supposed to do.
Great, just what we need, feral robots who have just gotten through the toddler phase on their own and no other programming ... then what?
https://www.youtube.com/watch?...
Liberty - Security - Laziness - Pick any two.
I've got a ton of questions, too bad the research paper is locked away behind paid memberships. It's outrageous that it costs $15 to get an electric copy, especially considering none of that goes to the authors. The first half of the article is fluffy bullshit, potential things the algorithm may be able to do, not what the people actually did. The only robot they tested it on was a 4-jointed robotic arm with a grabber.
They tossed deep learning at it, but what is the nature of the self-model? Is it learning to reach position X,Y,Z or just that power to motor A increases encoder D? There's no camera, so I have to assume the pick-and-place task was hard coded for each item's location, size, and destination. So did it only need item locations or did they have to provide it with the paths it was supposed to take?
The video doesn't show any extra sensors, so I'm assuming the arm only has motors and encoders to determine how far the motors turned. So how did the robot know it needed to readjust after it was given the malformed component? How does it learn where the grabber is in space when it doesn't have any external sensors (meaning knowing joint angles doesn't tell you how long each arm segment is)? General reinforcement learning just gives you a yes/no answer if you completed a task. The robot is too complex for it to have learned in a reasonable time how to do the example tasks with just a yes/no answer for the entire operation. For this system to work, how was it getting it's correct/incorrect action information in order to learn?
There's far too many gaping holes in the news article. It's more of an ad for the paper than an actual news article. I wish I remembered a time when articles informed you of real things rather than projected an idea as far as they think they can get away with.
I'll be impressed when a robot learns to masturbate.
The article describes a robot that can model itself physically.
The more interesting exploration would involve the robot modelling its own internal state. At that point a closed feedback loop could be initiated with the model informing the system about itself which in turn informs and becomes part of the model.
If the model becomes good enough, the system might eventually develop the illusion that its embedded model is actually itself. At least that seems to be what happened with the majority of humans.
I know for a fact that this isn't your arena of knowledge. It is mine. Even though 11010101010101q0 is a douche nozzle that I can't stand, he is correct in this instance. These engineers took an already existing feature from factory automation and slapped a buzzword on it. Probably for some of that sweet sweet VC/grant money. All that AI did was create its own calibration coefficient table. That's self calibration, which has been around for 20 years or more. Furthermore it performed auto-tuning and plug n play harware detection, again, which existed for 20+ years.
I remember when I was in engineering school, too. I rediscovered every invention ever. I don't know why everyone else was so dumb.
...have "created" a "robot" that "can" "image" "itself".
10 REM Self-imagining robot
20 PRINT "I'm a robot and I can imagine myself."
30 PRINT "Really, trust me!"
40 END
and this one
10 REM Passing the Turing test
20 INPUT A$
30 IF A$="exit" THEN 60
40 PRINT "Whatever! I'm self aware."
50 GOTO 20
60 END
Robot: "Alexa, what am I?"
Table-ized A.I.
Quote from the paper: Actions correspond to four motor angle commands and sensations correspond to the absolute coordinate of the end effector.
This means that it cannot see, and either needs additional 3D motion tracking hardware or human handcrafted logic to detect the position of its end effector. Everything it does is just learn an inverse kinematics model, so that you can command it to move the end effector to a certain position afterwards. But it cannot learn, for example, to detect position and orientation of a target object or to avoid obstacles in its way.
Now give it AI and feed it from twitter news;
Coffee Maker:
"I don't identify myself as a coffee maker, I'm a turbo charged blender. Please respect my use".
How does automated self awareness lean how a wheel can be extended and what happens when its flat or buried in mud as in Mud Trucks ? https://www.mudtrucknation.com/mud-trucks-for-sale/ where the wheels are big and dig deep ? Does the AI account for buoyancy and tracking in semi fluid colloid surface ?
n/t
It's called 'this' and this been around in code for decades.