Learn to Load the Dishwasher

We develop an algorithm to enable a personal robot to conduct one of the common tasks in our daily life – moving the objects to a specific place. One typical application scenario of this algorithm is to load a variety of kitchen items such as the plates, mugs, and bowls etc. into a dishwasher. For newly seen objects and their destination container (e.g. the dishwahser), our algorithm features a supervised learning procedure to decide both the position to grasp the objects and the locations to place them. The input of the algorithm includes the image captured by a camera and the point cloud data collected by a depth sensor. A few placing positions and the corresponding object’s placing orientations are selected by the trained SVM. Finally our method runs the point-based rigid-body simulation for each of the position/orientation candidates to pick the best placing strategy and proposes an placing order if multiple objects are considered.