• task-oriented
• grasp
• grasps
• pick
• place
• constraints

Task-oriented grasping refers to the problem of computing stable grasps on objects to allow subsequent execution of a task. Although grasping objects in a task-oriented manner comes naturally to humans, it is still very challenging for robots. Take for example a robotic arm employed in a industrial scenario. Nowadays, such a robot, in order to execute pick and place tasks, needs to receive objects in a known orientation and position. This is needed because the grasping action is pre-computed for that specific pose of the object, and it cannot vary to adapt to changes in the scenario. If the robot were able to reason about the best grasp to catch every object received, the objects could arrive in different orientations and positions. However, in order to do this, the robot should be able to infer the best way to grasp the object from the pose of the object and the required task. There are several challenges when it comes to this. First, the robot needs to know a set of different grasp candidates that can be used to pick the object, and be capable of choosing one of them. Second, it needs to know how to actually place the hand such that the task can be executed with respect to the environmental constraints. Finally, algorithms for task-oriented grasping should be scalable and have high generalizability over many object classes and tasks.
In this thesis, we present a novel method for task-oriented grasping that rely on optimization. The task considered as illustrative of the category of tasks taken into account is a pick and place, where an object needs to be moved from one place to another in a constrained environment. The choice of such a task is due to the fact that it is similar to tasks like packaging and assembly, common tasks in industry. The idea this work proposes is to select within a set of grasp candidates the best grasp that allows to accomplish the whole task, taking into account the initial pose of the object, the final pose of the object, and the environmental constraints that prevent some grasps from being feasible.