Caging grasps would handle objects without full immobilization and enable dealing with the object`s position and orientation uncertainties. Most previous works have constructed caging sets by using the geometric models of the object. This work aims to present a learning-based method for caging objects only with its image. A multi-task learning method is developed for caging grasps, where the caging region is directly learned from the image of the object. Furthermore, several different caging tasks are trained with a joint model using the sample data of the tasks. Therefore, we could avoid the collection of plenty of training samples. Simulations show the validity of the method.
展开▼
