An RI Group usually consists of members from multiple labs here at The Robotics Insitute as well as other labs at the univiersity.
Our group is interested in inferring physically and spatially grounded representations from perceptual input, and leveraging these for advances in fundamental problems in computer vision and robot manipulation. We believe that to enable machines to understand the physical world, we should reduce the reliance on supervision by annotation, and instead develop learning mechanisms informed by the real, physical world we live in – by incorporating our knowledge about its structure and laws as a ‘meta-supervisory’ signal.
