In this video, we consider the task of a personal robot organizing a room by placing objects stably as well as in semantically preferred locations. While this includes many sub-tasks such as grasping an object, moving to a placing position, localizing itself and placing the object in a proper location and orientation, it is the last problem — how and where to place the objects — that is our focus in this work and has not been widely studied yet. We formulate the placing task as a learning problem. By computing appearance and shape features from the input (point-clouds) that can capture the stability and semantics, our algorithm can identify good placements for multiple objects. In this video, we put together the placing algorithm with other sub-tasks to enable a robot organize a room in several scenarios, such as loading a bookshelf, a fridge, a waste bin and blackboard with various objects.
展开▼
