Slashdot Mirror
New 'Deep Learning' Technique Lets Robots Learn Through Trial-and-Error
jan_jes writes: UC Berkeley researchers turned to a branch of artificial intelligence known as deep learning for developing algorithms that enable robots to learn motor tasks through trial and error. It's a process that more closely approximates the way humans learn, marking a major milestone in the field of artificial intelligence. Their demonstration robot completes tasks such as "putting a clothes hanger on a rack, assembling a toy plane, screwing a cap on a water bottle, and more" without pre-programmed details about its surroundings. The challenge of putting robots into real-life settings (e.g. homes or offices) is that those environments are constantly changing. The robot must be able to perceive and adapt to its surroundings, so this type of learning is an important step.
This seems more like basic-level stuff... learning from your mistakes. That strikes me as the sort of thing that would be "hardwired" in everything from nematodes to primates. Why is this news?
XML is like violence. If it doesn't solve your problem, you're not using enough of it. --AC
Cue the Skynet / Matrix references in 3...2...1...
/. zen: Imagine a Beowulf cluster of Beowulf clusters...
... only expendable human coworkers
There is a Genetic Algorithms textbook from 1989 that covers generational learning and "mutating" the parameters until you get to the end state in the best way possible. My AI knowledge isn't great but I wouldn't be surprised if there are ideas that pre-date the '89 text.
Does anyone know what the software controlling the robot is doing under the hood that's different?
Good, now you can clean up my messes! - George J.
"algorithms that enable robots to learn motor tasks through trial and error"
1) Do random thing x out of n possible with probability p_x=1/sum(p1,p2,...pn)
2a) If reward then p(x)=p(x)+k
2b) If no reward then p(x)=0
3) GOTO 1
They are becoming the 2015 equivalent of "Frist Post, or "Welcome from the Golden Girls".
The shepherds did so well protecting the flock that the sheep no longer believed that wolves existed.
I really recommend these two books by Sladek: http://en.wikipedia.org/wiki/R... they're very funny satire about a naive, learning robot in a cruel illogical world. This is what our little friend here can expect.
On y va, qui mal y pense!
Why in the world would you build a benevolent God ... when you could become God?
We won't make the machines take care of us. We will augment our abilities by hooking our brains up to the machines directly.
Those of you who don't do this might wind up being "take care of" in some historical wildlife preserve somewhere. The rest of us will travel the stars, the limitless depths of virtual space, and other frontiers of discovery that we cannot today imagine.
Reminds me of http://www.newscientist.com/ar... from 2002. Robots goal was to raise its altitude without knowing its actuators ahead of time.
There's nothing new about this. Here's the important section of the article:
BRETT takes in the scene, including the position of its own arms and hands, as viewed by the camera. The algorithm provides real-time feedback via the score based upon the robot’s movements. Movements that bring the robot closer to completing the task will score higher than those that do not. The score feeds back through the neural net, so the robot can learn which movements are better for the task at hand
All it is is reinforcement learning and a (deep?) neural network. That's how you're supposed to do it. This is entry level AI applied to an expensive robot with some marketing (it's a press release not a research paper).
As the PR2 moves its joints and manipulates objects, the algorithm calculates good values for the 92,000 parameters of the neural net it needs to learn.
With this approach, when given the relevant coordinates for the beginning and end of the task, the PR2 could master a typical assignment in about 10 minutes. When the robot is not given the location for the objects in the scene and needs to learn vision and control together, the learning process takes about three hours.
Abbeel says the field will likely see significant improvements as the ability to process vast amounts of data improves.
So now we see the real results. The only improvement came from faster hardware.
What they didn't demo was telling the robot to repeat what it had learned and apply it to something. If this was released as it, it would take days? to do anything in the real world and someone would have to mentor it saying "yes" or "no" to every tiny movement it performed (there's an operating system built on this principle, no one uses it). This would be slightly impressive if it reused what it learned (could apply case based reasoning or something else), but it doesn't. To be fair the students did a good job implementing something that was already known to work and you need to do that before you can take the next step to something new, but this isn't that step. I'm a Masters student focusing on real-time strategy (RTS) AIs. All these things have been done before, read the research papers. What hasn't been done are public demos. There are fewer and fewer RTS games and the companies don't want to spend the CPU time to use any of the more advanced AI algorithms. The research is advancing, the products are not.
Better known as 'learning' to everyone not trying to exaggerate an claim of artificial intelligence.
It's excellent progress, which is why I don't think it should be watered down by being compared to the simple algorithms.
People were doing this when I was an undergrad, almost 20 years ago. I specifically remember a six legged robot that had to figure out how to walk by itself.
Help you put in your contacts sir? ...!!! ...!!!
*adjusting servo*
Help you put in your contacts sir?
A robot that can replace the Salvos volunteers at hanging up donated clothes.
Self-correcting curious entities. If there's teeth on them, we're bound to become enemies.