Google researchers have developeda robot capable of learning to walkthrough different environments alone, without the need for a human to intervene, thanks to Artificial Intelligence (AI), and with the aim of making them more useful in the real world.
Autonomous robots are already a reality, but it is not yet possible for them to learn autonomously. So far robots use reinforcement learning algorithms to learnnew movements through trial and errorthat depend greatly on human intervention. For example every time the robot falls, someone has to go pick it up.
However, Google has managed to get a four-legged robot to learn to walk forward and behind or to the left and right by itself using the latest advances in algorithms, as reflected by the Massachusetts Institute of Technology (MIT), in the United States, in their website.
The research is based on another study from the University of Berkeley (United States) that looked for formulas for a robot to learn on its own in the real world. Generally, reinforced learning takes placeusing a simulationwith a double of the robot inserted in a virtual stage to learn how to behave.
Once the algorithm has enough information to function safely, the algorithm is introduced to the physical robot. This technique prevents the robot and the surrounding environment from suffering any type of damage while testing.
Trained in the real world
For this new investigation, they decided to train the robot directly in the real world. For this they had to develop a more efficient algorithm thanless evidence neededand therefore less mistakes. Also because the environment presented variations, the robot was able to adapt to them.
However, the robot still required human intervention, so they had to carry out a series of changes. First, they delimited the terrain that the robot could investigate and the maneuvers it could do. They also limited their movements in addition to making it more cautious in order to minimize damage each time the robot fell.
With these changes, the researchers got the robot to learn to walk through different environments alone. This work greatly expands the possible activities that robots will carry out to move through different environments without human presence.