Injured robots learn to limp (1:32s video)

[Perspicuo]

Injured robots learn to limp

 

[none]

that is totally cool for now....

 

[Perspicuo]

That's behavioral adaptation. Real complex neuro animal stuff. I don't know what they're using for proprioception, though.

 

[barbos]

They said it choose that style of walking from thousands options researchers (!!!) programmed into it.
So it was already there.

 

[EricK]

They said it choose that style of walking from thousands options researchers (!!!) programmed into it.
So it was already there.

Unless what they mean is that there are different ways each leg, or joint, can move and these are programmed in, and the robot tries various combinations of those (which would number in the thousands) until it stumbles on one which works.

 

[barbos]

I don't know what they mean exactly but I am not impressed either way.
Learning to limp is not fundamentally different from leaning to walk on perfectly working legs.

 

[fromderinside]

Gee. Another demonstration of a program that adapts to change. What does one say about a machine vision system that shuts down a process when it 'detects' something that is out of spec? ohIdunno ...... a program that learns to limp?

 

[RavenSky]

Clearly I'm the only one here that agrees the scientist is a big meany for hurting the robot . (& I apologize to the Roomba when I trip over it in my house)

But yes, that was very cool. Thank you for posting it!

 

[Jimmy Higgins]

I don't know what they mean exactly but I am not impressed either way.
Learning to limp is not fundamentally different from leaning to walk on perfectly working legs.

Robot programmed to walk without all legs, "learns" to walk without all legs.

 

[thebeave]

Kinda reminds me of the end of Terminator, when first the Terminator cyborg crashes his motorcycle, then walks with a limp. Then, his outer "skin" fries off in a fiery truck crash, so he walks out of it with just his metallic skeleton. Then, he loses his lower half and starts crawling around with just his arms, before he's finally destroyed.

It won't be long folks.

 

[fromderinside]

So long?

 

[ApostateAbe]

I expect that the future of robotics would be something akin to natural selection, but, instead of robots making copies of themselves, they upgrade themselves, choosing the most effective out of any of many pseudorandom actions, to do any task more effectively. A new robotics language may be appropriate, maybe something akin to the language of DNA, to make random variations more likely to produce positive results.