Self-aware robot adapts to injury

[PeterD]

From the January '07 Machine Design magazine:

Researchers at Cornell University have built a four-legged robot that changes its programming to adapt to injury.The underlying algorithm, according to the team, could be used to build more complex robots that can deal with uncertain situations, like space exploration, and may help in understanding human and animal behavior.

The robot works out how to control itself in a process similar to the way human and animal babies discover and manipulate their bodies. The ability to build this "self-model" is what makes it able to adapt to injury.


You can read the complete article here. (http://www.machinedesign.com/ASP/viewSelectedArticle.asp?strArticleId=61809&strSite=MDSite&Screen=CURRENTISSUE&CatID=3)

[HarryBotter]

This is interesting stuff. Is there any consensus out there on the best way to model learning capabilities in robots yet?

Harry

[TheDuck]

This reminds me of a demo I saw with, I believe, one of the Tilden-based robots. He would keep bending the legs of the poor six-legged thing into many ridiculous configurations and it would keep "figuring it out" and correcting its locomotion. Since most of Tilden's creations are ingenious but really straightforward implementations I'm sure there was some equally straightforward (and ingenious) that allowed this to happen.

Having said that, I think the success of such an endeavour would depend highly on something that has been missing from robots: FEEDBACK. I must admit I'm seeing more of it but it really must be used as a constant input to EVERYTHING the robot does. When you tell the motor to run for 90 revolutions you cannot expect the robot has turned 90 degrees. Rather you need to perform an action until you have definitive feedback that the action has been performed. Once you've done this you swing wide a door to more possibilities. For example, should you not be getting the feedback you expect then you need to try other things. By "other things" I mean you introduce increasing levels of randomness into your outputs and monitor to see if any of these get you any closer to the feedback input you are expecting. If there is still no change you are either REALLY stuck or you can no longer rely on your input. This is a bit harder I would think. Imagine waking up to find you are blind. You still adapt but that's because there are patterns of recognition from your other senses that help you. If you can make a robot model multiple feedback patterns to represent goals then I presume you are at the Santa Fe Institute, MIT, Carnegie Melon, .... ;) (P.S. I'd REALLY like to chat with you! :D )

Ok, I'll stop there. Hopefully you aren't expecting to give me any change for that two cents. But you DID start me on one of my favourite topics!

[Rover]

Thanks Pete, good article.

[dporter]

This reminds me of a demo I saw with, I believe, one of the Tilden-based robots. He would keep bending the legs of the poor six-legged thing into many ridiculous configurations and it would keep "figuring it out" and correcting its locomotion. Since most of Tilden's creations are ingenious but really straightforward implementations I'm sure there was some equally straightforward (and ingenious) that allowed this to happen.



Your right, Tilden has made several amazing robots. What is surprising is that they use very basic parts (capacitors, transitors, diodes) often from junk, but still have complex behaviors. From what I remember from reading about his robots,(and hopefully someone will correct me if I explain something wrong) they are analog and not digital and the circuit gets feedback from the change in voltages from loads on the circuit like from motors. So as one motor gets stuck or strains against an object the circuit responds to the voltage changes by shifting current to the other motors. That really simplifies what is going on but there is a lot of information about BEAM robots and circuits at http://bftgu.solarbotics.net/starting.html if anyone would want to learn more about how they work. I like BEAM robots because they are relatively low cost, but can be as complex as your creativity will allow.

[hgordon]

There is a very interesting book on this subject called "How the Body Shapes the Way We Think" - http://www.amazon.com/How-Body-Shapes-Way-Think/dp/0262162393/ -
by Rolf Pfeifer and Josh Bongard (author of the article that Peter flagged). The area of research which encompasses these developments is often referred to as "embodied cognition" or "embodied intelligence", and it combines many disciplines of robotics research and artificial intelligence.

There are very good online interviews of Rolf Pfeifer here -
http://lis.epfl.ch/resources/podcast/2007/02/rolf-pfeifer-new-ai.html

as well as Josh Bongard and Hod Lipson here -
http://lis.epfl.ch/resources/podcast/2006/12/josh-bongard-and-hod-lipson-resilient.html

[HarryBotter]

Thanks Howard, looks like a great book, I've ordered it.

Harry