In this project we are developing deep learning methods for robotics. Deep learning is a branch of machine learning that is concerned with learning structure, representations, and underlying patterns in complex “raw” data, such as images and sounds. Traditionally, these methods have been applied to passive tasks, like recognizing cars or pedestrians in camera images, where the method is “trained” by presenting it with hand-labeled examples of the objects that must be recognized. In this project we aim to develop deep learning techniques that can be deployed on a robot to allow it to learn directly from trial-and-error, where the only information provided by the teacher is the degree to which it is succeeding at the current task.
More information can be found at rll.berkeley.edu/deeplearningrobotics.
Related News: “New Approach Trains Robots to Match Human Dexterity and Speed” (The New York Times, 5/21/15)
Image credit: Peter Earl McCollough for The New York Times