My PhD research focuses on task-oriented grasping from sensor data. There are a few key aspects to this problem: the first one is the mathematical definition of a task, the second key aspect is using this definition of a task, how do we formalise the notion of grasping to perform a task, and the last part is incorporating noisy sensor data in the form of partial point clouds from an RGBD camera. In our work, we define a task as a constant screw motion or a sequence of constant screw motions to be specific. Using this definition of a task, we have developed a task-dependent grasp metric as a second-order cone program which evaluates the feasibility of object-robot contact locations to impart a specific constant screw motion. We have developed algorithmic approaches which use this notion of a task and a grasp to perform grasp synthesis on point cloud data using its simplest possible geometric representation i.e. a bounding box.

Related Papers: