To learn about their world, children play with toys, moving and manipulating objects without any sort of training. Now, researchers led by Sergey Levine, assistant professor of electrical engineering and computer sciences and member of the Berkeley Artificial Intelligence Research Lab, have developed technology that enables robots to do the same: manipulate objects they have never encountered before by building on previous play and observations. Using this technology, called visual foresight, the robots can predict what their cameras will see if they perform a particular sequence of movements. Although these predictions extend only several seconds into the future, they are enough to allow the robot to move objects around on a table without disturbing obstacles. Crucially, the robot can perform these tasks without any help from humans or prior knowledge about physics, its environment or the objects. Through this, the robot builds a predictive model of the world, which it uses when it encounters new objects. In the future, this technology could produce more intelligent robotic assistants in homes and help self-driving cars anticipate events on the road.
Reach the editors at firstname.lastname@example.org