Visual Competencies for the iCub

Video summarizing the work achieved in order to allow a humanoid robot, the iCub, to develop improved exploratory visual competencies. It presents a framework that combines state-of-art boosting techniques (an online approach implementing the Multiple Instance Learning paradigm within a boosting framework) merging the information recovered from the cameras with the multisensory information provided by the embodiment on a humanoid robot. This permits the robot to learn and store the visual appearances of objects in order to perform precise goal directed tasks using complex behavioral, cognitive skills.

Credits: Carlo Ciliberto, Ugo Pattacini, Vadim Tikhanoff and Stephane Lallee.