For years, researchers and scientists have tried to design robots that were and acted in the most similar way to what a human would do, the enormous advances in technology and Artificial Intelligence have brought us very close to this goal, but it is that this new technique could completely change how these machines work in a few years.
Since the first robots, the way their programmers have had to teach them to do something was to go to the computer and insert a series of commands and codes for the machine to translate and carry out the action.
In recent years we have seen new techniques driven by Artificial Intelligence (AI) that seek to involve the robot in the learning process and not be the human who programs it so that it knows how to do it directly. And that is precisely what the team of researchers from Carneige Mellon University (United States) has done, designing a new method in which systems observe people perform an action in order to learn how to perform it themselves.
With this algorithm, called the learning of human imitation robots in the wild (WHIRL for its acronym in English), the researchers ensure that the system is capable of learning to carry out an action just by watching a video tutorial.
“Imitation is a great way to learn,” Shikhar Bahl, a PhD student at the Robotics Institute, said in a statement. “Getting robots to actually learn by directly observing humans remains an unresolved problem in this field, but this work takes an important step in enabling that capability.”
imitate to learn
In the end, this algorithm is intended to work a bit like a baby’s brain does, that is, it watches what they are doing and repeats it back. The learning process can last several hours, there is no set time and it will only end when the machine has successfully completed the task and is in control of it.
And unlike what many people do, the system tries to find a way to perform a more convenient action with its limitations. In the examples given by the researchers, domestic tasks such as picking up the garbage, washing the dishes, opening a fridge are shown, where in each of them the robot has had to analyze the movements of the human and apply them within its restrictions to be able to carry them out. cape.
Undoubtedly, this advance makes robots more and more intelligent, mainly because they act more and more similarly to people, it is no longer programming a robot to do something, but teaching the robot how to do something which will presumably lead to reducing these machines from failing because a bug has been inserted into their system.
Facebook Comments
Disqus