BBC: A new robot design uses a trial-and-error algorithm to diagnose problems and find a way for the robot to keep operating. The algorithm was modeled on the behavior of injured animals, which have been known to adapt to a wide variety of injuries. Unlike current robot designs, which rely on a limited number of fixes preprogrammed into their system, the new design can handle unforeseen problems by allowing robots "to creatively discover compensatory behaviors,” say the researchers in their paper published on the arXiv website. Their prototype robot was successful in learning to adjust after having one of its six legs broken, and it did so in just two minutes. Such a self-learning robot could be useful in situations where humans can’t go, such as certain space missions, search and rescue operations, and disaster response.
