AI Leg Teaches Itself to Walk
Researchers at the University of Southern California have developed a robot leg that corrects itself when it trips, when it was never programmed to do so.
The nature inspired robot uses a complex algorithm in which it can teach itself a different walking style only after 5 minutes of unstructured play.
It can then develop upon the walking style, analyse its surroundings and then adapt to new, different walking styles without the need for anybody adding specific code to give it that extra functionality. This breakthrough in research could lead to a development and advancement in the making of prosthetics and robots that are able to efficiently interpret, analyse and adapt quickly to changes in environment. This means that these robots will be suitable for use in different terrains whether it be the dessert, grass, snow, etc.
Essentially, the robot had learned to mimic the walking style of human babies by being allowed to explore and analyse its environment through free play or what is known as ‘motor babbling’.
Flexible and adaptable prosthetics like these could be further developed by implementing them into robots that can be used for space exploration, where terrains are very much different to that of our planet.
As time progresses, more money is invested into the research, understanding and wider implementation of artificial intelligence. Not only would this type of technology be useful for robots traversing through different terrains but another suitable application could be in providing alternative and better medical prosthetics than the ones currently available. The ability for robots to learn habits and adapt to a certain style could have a large impact on prosthetics as a whole. Imagine a leg prosthetic that learns and is able to mimic the movement of its owner.