Robots learn to move themselves

[waqo]

The power of neural networks is clear for all to see in this remarkable development as far as programming and the future of AI is concerned.

If this can be implemented and sustained then neural networks could be used to realistically simulate real life (given the processing power)!

Source: BBC NEWS | Technology | Robots learn to move themselves

Researchers in Leipzig have demonstrated software designed for robots that allows them to "learn" to move through trial and error.
The software mimics the interconnected sensing and processing of a brain in a so-called "neural network".
Armed with such a network, the simulated creatures start to explore.
In video demonstrations, a simulated dog learns to jump over a fence, and a humanoid learns how to get upright, as well as do back flips ......

......."In the beginning, we just drop a robot into a space. But they don't know anything, so they don't do anything," Professor Der said. The neural network eventually picks up on electronic noise, which causes small motions.
It eventually tries larger motions as it learns about its range of movement. "It's like a newborn baby—it doesn't know anything but tries motions that are natural for its body. Half an hour later, it's rolling and jumping," Professor Der said..........


......

Moreover, the software can be used with any kind of robot, and Professor Der has tried the system on simple wheeled systems. "I call it a plug-and-play brain," he said.
"The classic thing in robotics is 'bring this' or 'play this chess game and win'—the task is given," says Daniel Polani of the University of Hertfordshire. "Ralf Der's system is only defined by what it perceives and does, but there's no goal. It's a very good approach."

[X2thaU]

woah bro we don't need any geth invasions any time soon

[waqo]

Mass Effect is an awesome game and really this is potentially a very basic step towards a geth type existence being a realistic possibility.

Awesome thing is that neural networks work on many different levels they can not only optimize a function universally but can find local optimums and then in the next loop utilize the evolved information. This allows them to quickly hone in on a solution similar to how our brain works!

[GaaranGaara]

This sounds very cool. Soon there will be no need for us humans and the robots will take over the world!

No, but seriously we humans better not develop AI too far. Just because if robots soon are able to do our jobs. The level of fitness would drop even more. Humans would just be lazy slobs.

[Gathan]

Originally Posted by GaaranGaara

No, but seriously we humans better not develop AI too far. Just because if robots soon are able to do our jobs. The level of fitness would drop even more. Humans would just be lazy slobs.

With a neural network we don't need to develop the AI, it develops itself.
And will evolve into the supreme robotic being

[Rafcio]

Imagine Quake 3 bots that learn to play the maps.

[Ericgamer1]

That's awesome. The biggest reasons why machines have been unable to add much outside of what we program them to do is that they can't make decisions based on what they're not told to do. Pretty big leap forward.

[Ycagel]

Now, if they could just stop the robots from forgetting everything they learn as soon as they learn it, this could be awesome. To see that simulated dog struggle for like an hour on getting over a fence, just for it to have to spend another hour on another identical fence because it forgot how to get over one... Yeah, must have been annoying.

[Gathan]

Originally Posted by Ycagel

Now, if they could just stop the robots from forgetting everything they learn as soon as they learn it, this could be awesome. To see that simulated dog struggle for like an hour on getting over a fence, just for it to have to spend another hour on another identical fence because it forgot how to get over one... Yeah, must have been annoying.

Obviously you don't get the concept of learning, in other words it would remember how it got over the last fence and use the same procedure or learn how to modify what it already learned to get over a slightly different fence. So no it wouldn't spend another hour getting over a identical fence because it would already know how to do it.

[San Goku]

Even better for space exploration if this can pick up really fast.

[Gathan]

Originally Posted by Rafcio

Imagine Quake 3 bots that learn to play the maps.

Ya that would be pretty fun, that is up until the point the bot learns how to play the maps like a gaming god with perfect aim

[Ycagel]

Originally Posted by Gathan

Obviously you don't get the concept of learning, in other words it would remember how it got over the last fence and use the same procedure or learn how to modify what it already learned to get over a slightly different fence. So no it wouldn't spend another hour getting over a identical fence because it would already know how to do it.

Yeah, uh, no. Here's something from the site Waqo linked to.

For now, the network learns behaviours such as how to stand up, but promptly forgets them. Der and his colleagues are working to create a long-term memory, so that when the robot finds itself in similar situations, it knows what to do

Dunno why that part isn't in the first post. It just say "similar situations," but when it forgets how to stand, I don't see it remembering how to get over a fence.

[Gathan]

^Oh that sucks, well I guess are supreme robotic overlord is just going to have to wait for them to create memory for it.
Still making this first step is impressive progress and overcomes the need to write programming code at all.

[o({})o]

Neural network programming has been around for ages... if comptuers can learn to conjugate irregular verbs and do advanced calculus, then basic movement is not worthy of being called news.

[waqo]

Originally Posted by o({})o

Neural network programming has been around for ages... if comptuers can learn to conjugate irregular verbs and do advanced calculus, then basic movement is not worthy of being called news.

the concept has existed for ages, however the key for me is that this bit of code can be so universally adapted. It's like provide the software with a body and muscles and it shall move.

The memory implementation is a hurdle but I think things like movement are much more difficult for 'intelligent computers' to handle than recognizing verbs and calculus optimization.

Edit: I forgot to mention something that is also quite unique about this, it says that this system takes about a half an hour to optimize towards actions like rolling and jumping, previous incarnations of bipedal movement have required many many generations to arrive at results like that, reducing practicality. If the programmer has found a method of quick evolution for the neural networks rather than honing in on random optimized mutations (all time frames I have previously seen for comparison have been in terms of many hours, days or even months) then I would say it is newsworthy.

[Cane_The9lives]

Could this be considered self awarness at the most fundemental of levels?

[San Goku]

Thats what i'm getting from this thread, or else it wouldn't make sense making a thread like this.

[Cane_The9lives]

Originally Posted by San Goku

Thats what i'm getting from this thread, or else it wouldn't make sense making a thread like this.

Now we need a time machine and john connor.

[Righteous Judgment]

Originally Posted by Ericgamer1

That's awesome. The biggest reasons why machines have been unable to add much outside of what we program them to do is that they can't make decisions based on what they're not told to do. Pretty big leap forward.

in enslaving mankind dun dun dun

[BlitzKid623]

The technology really is improving everyday. I've seen a few Youtube videos which present some of the technology. There was one robot who was able to move around obstacles, and learn from his mistakes.

Japan has some crazy stuff going on over there.