Media

Videos

PR2 robot throwing a ball
The video demonstrates a reinforcement learning framework that is used to train a deep neural network architecture to acquire skilled sensorimotor behaviors, such as ball throwing. The trained neural network architecture maps raw RGB input images to a trajectory of motor activations to complete the given task. The approach is described in this article.
PR2 robot together with a human controlling a ball on a plank
In this video, we present a data-efficient reinforcement learning framework that is applied to learn physical collaboration with a human partner. A PR2 robot learns controlling a ball on a plank, jointly with a human partner, given visual and haptic sensory observations.