Robots that learn by observing and imitating human actions

Until now, when the programmer wanted the robot to do something specific, he had to do the following: open the computer, click on the appropriate program, and generate the lines of code needed to translate that intention into a language the automaton could understand.

But what if robots could sit down and learn from us? Yes, just like children learn to walk by watching their parents, robots can do the same when it comes to learning new movements. And let’s leave the conventions, since this has been a reality for some time.

And although imitation and reinforcement learning are the two most popular methods today. The first involves controlling a robot to teach it how to perform a task, and the second involves training a system with millions of images.

Several researchers are exploring an even more intuitive method that effectively trains a system by watching a human perform a task.. A team at Carnegie Mellon University has created the WHIRL algorithm that can train a system by watching a video.

In their demonstrations everyday mobile robotic arm learns to perform more than 20 everyday tasks such as opening and closing drawers, equipment and take out the trash.

“Imitation is a good way to learn“, say its creators.”Getting robots to actually learn by directly observing humans remains an unsolved problem in the field, but this work takes an important step towards realizing this ability.“.

It’s easy to see how a feature like this could be especially useful in the home, when roboticists envision such systems one day being deployed to help the elderly and others with mobility issues.

In the case of WHIRL, no special plugins are required. The robot simply tries to complete a certain task until it succeeds, even if it takes hours to fully master it.

According to the university, your own approach may not be identical to the human one, rather the system is looking for the best way to complete the task based on your own hardware limitations.