iCub learns objects

This video demonstrates an architecture allowing an humanoid robot iCub to learn new objects presented to it. This is done by adding the concept of object to an existent low-level attention system of the robot. When the robot first encounters an unknown object, found to be within a certain (small) distance from its eyes, it stores a cluster of the SIFT visual features present within an interval about that distance, using depth perception. Whenever a previously stored object crosses the robot's field of view again, it is recognized and mapped into an egocentrical frame of reference and gazed at. This mapping is persistent, in the sense that its identification and position are kept even if not visible by the robot.

More info at http://mediawiki.isr.ist.utl.pt/wiki/Cognitive_Architectures

Category:

Science & Technology

Tags:

• iCub
• humanoid robot
• cognitive architecture
• SIFT
• egosphere

[Explorer360]

Advertising coco pops ftw?

[cdoep]

Wow

[kkuznecovv]

Creeepy......

[wlorenz65]

@WurmD Well, there is a project at IM-CLeVeR.eu but I haven't seen any videos of the CLEVER-K iCub yet.

One thing is clear: A true learning robot needs a skin. This makes it easy to filter out all those AI cheaters :)

[WurmD]

@wlorenz65 the human brain's programmer took a little longer ^_^ (a few million years), and the human robot takes a little longer to enact learning :) (a few years)

But we'll get there ^_^, inevitably

(unless we kill ourselves XD)

[roleplayman]

@noxure there is already neural network that classifies things

[Anything2bCool]

ketchup ...catsup...ketchup ...catsup...

[Zormac]

After it has learned everything, you will be amazed as it doesn't move and flicks its head around like the exorcist, trying to gaze upon everything.

[RebleINR]

wlorenz65, yeah that's classified as a toy seriously, a Robot is automatic... what is that? the only thing I see that robot do on it's own is see the object and turn its head. I'd say it's 1.8% complete robot.

1 Point for making it turn its head and

0.3 for making it detect something and then the rest for memorizing it.

[wlorenz65]

Human brains find out on their own that it is useful to separate the

enviroment into objects.

iCub instead needs programmers to implement the concept of an "object" in its software.

This means that iCub cannot learn to recognize combinations of objects or procedures as useful. It needs programmers again.

[noxure]

Nice. Now try to figure out how it can understand the "concept" of a cereal box, instead of memorizing every individual box. We humans know thousands of objects and we can recognize stuff we've never seen before too.

We need better processors, fast. Can't wait for those groundbraking memristor chips to come out. Neural network processor units... NNPU's. :)