Google creates a robot that is capable of learning to walk alone

Image result for Google creates a robot that is capable of learning to walk alone

A set of Google engineers has been able to create a robot that does not require human help, and can only learn to perform different tasks, such as walking, based solely on a previous algorithm about real environments.
Most of the robots that exist today require the support of an engineer, and extensive programming work to provide an algorithm that has thought about each of the situations they can face, but from Google, they want to go further. This is an investigation that began long ago, first developing artificial intelligence under a virtual environment where the robot was learning how to move. However, these types of environments mean that not being specifically the real world, they can end up causing machine failures.
With the experience gained, the researchers trained a robot from the beginning in the real world by first designing a more efficient algorithm that was able to learn with fewer tests and fewer errors. For this, they had to use different techniques so that the robot could perform multiple things at the same time but without falling, learning from the failures and introducing the solution later in the algorithm.

Robot Google

As the research team accumulated adjustments in the robot's algorithm, the machine was able to walk itself through different surfaces, including flat ground, a doormat with cracks and a foam mattress.
With this learning technique for robots, these robots could be trained in real environments so that they could move to places where they usually cannot. Assistant professor Chelsea Finn of the research team said that "removing the person from the learning process is really difficult."
"By allowing robots to learn more autonomously, they are closer to being able to learn from the real world in which we live, rather than in a laboratory as originally done," he adds.

Image result for Google creates a robot that is capable of learning to walk alone

Still, there is a long way to go, and for the future researchers plan to adapt their algorithm to different robots or even several so that they can learn at the same time under the same environment.

Post a Comment