Iowa State Robot Learns To Learn
Writing by Evan Ackerman on Wednesday, 29 of July , 2009 at 5:16 am
This unnamed humanoid (or “uh” for short) from Iowa State University’s Developmental Robotics Lab is trying to figure out how the world works in the same way as a 2 year old does: by attempting to break stuff. Well, not exactly break stuff, but the robot does manipulate objects in a variety of different ways to try to establish their physical characteristics. It pushes things, shakes things, and drops things, and gradually builds a mental picture of sorts of what kind of object its playing with.
This type of experimentation is a distinctly biological method of learning, and has the advantage of not relying to heavily on one type of sensor… Robots tend to focus on visual information (in some wavelength or other, anyway), but when a human gets introduced to a new object, we’re more inclined to grasp it and turn it over in our hands while looking at it. We then store all of the characteristics of the object, and if we encounter something similar, we can make inferences about what other characteristics it might have. Unnamed humanoid is pretty good at doing the same thing, using what it’s figured out about a set of objects to identify them with an accuracy of 99%.
The next step that’s currently being worked on is experimentation with a goal in mind. For example, the robot is given a pencil and has to figure out how to draw something. If you later give the robot a pen, it should be able to determine that the pen is much like the pencil and use it the same way. And differences are equally important to discover, since differences generally provide important information about specific functions. Eventually, researchers hope to create some kind of robot butler that’s able to teach itself new tasks largely without your help.
There’s a video after the jump, with some decent footage that’s almost, but not quite, entirely spoiled by a TV host.
[ Iowa Developmental Robotics Lab ] VIA [ Plastic Pals ]
Comments (2)
Category: Artificial Intelligence,Research
- Add this post to
- Del.icio.us -
- Digg
Comment by Alex
Made Wednesday, 29 of July , 2009 at 3:50 pm
Gizo Watch got it from my site. Would be nice if you could link back to me (2nd time you’ve overlooked me as a source). I just started the site, and I could use the exposure. Thanks!
Comment by Evan Ackerman
Made Thursday, 30 of July , 2009 at 12:13 am
Sorry about that, I’ll try to be more careful.
