Researchers at the University of California, Berkeley have developed a robotic learning technology that enables robots to foresee the future.
This robotic learning technology called “Visual Foresight” allows robots to imagine their future actions in order for them to manipulate things they have not come across before. This technology, the UC Berkeley researchers said, could help in producing more intelligent robotic assistants in homes as well as produce autonomous cars that could foresee future events on the road.
Visual Foresight’s capability is fairly simple for now, with predictions limited only to several seconds into the future. This, however, is sufficient for the robot to figure out how to move objects around on a table without colliding with any of the obstacles.
According to the UC Berkeley researchers, their robot learns to foresee the future without help from humans and prior knowledge about physics, what the objects are and the environment.
The researchers said the robot learned to manipulate objects similar to humans – learning things along the way through interactions with a variety of objects during their lifetime. This is similar to learning by playing in the case of a child.
After this play phase, the researchers said, the robot then builds a predictive model of the world. The robot uses this predictive model to manipulate new objects that it has not encountered before.
At the core of UC Berkeley’s robotic learning technology is a deep learning technology based on convolutional recurrent video prediction, known as dynamic neural advection (DNA) – a model that predicts how pixels in an image will move from one frame to the next based on the actions of the robot.
“In that past, robots have learned skills with a human supervisor helping and providing feedback. What makes this work exciting is that the robots can learn a range of visual object manipulation skills entirely on their own,” said Chelsea Finn, a doctoral student in Levine’s lab and inventor of the original DNA model.