On the generation of a variety of grasps

Abstract

In everyday life, people use a large diversity of hands configurations while reaching out to grasp an object. They tend to vary their hands position/orientation around the object and their fingers placement on its surface according to the object properties such as its weight, shape, friction coefficient and the task they need to accomplish. Taking into account these properties, we propose a method for generating such a variety of good grasps that can be used for the accomplishment of many different tasks. Grasp synthesis is formulated as a single constrained optimization problem, generating grasps that are feasible for the hand's kinematics by minimizing the norm of the joint torque vector of the hand ensuring grasp stability. Given an object and a kinematic hand model, this method can easily be used to build a library of the corresponding object possible grasps. We show that the approach is adapted to different representations of the object surface and different hand kinematic models.

Dexterous grasping under shape uncertainty

“…Due to the high dimensionality of the problem, the optimization is conducted in a projected space of lower dimensionality using hand synergies [13] or eigen grasps [14]. There are also works that formulate the optimization in the original hand configuration space [15,16]. However, this is computationally expensive and the obtained grasps are hand-dependent.…”

“…is the i-th primitive contact wrench of the j-th contact point [16], which depends on the contact points, contact normal directions and friction coefficient. Note that in [55] this property is also defined as force closure, here we adopt the definition in [54] where force closure additionally requires that the internal forces are controllable by the hand.…”

On computing task-oriented grasps

“…In our previous work [11], we proposed a method for generating a variety of grasps by formulating grasp synthesis as a single constrained optimization problem. The computed grasps were feasible for the hands kinematics and were obtained by minimizing the norm of the joint torque vector of the hand ensuring grasp stability.…”

“…We have shown in our previous work [11] that Interior Point OPTimizer (IPOPT) method proposed by Wächter and Biegler [22] is adapted to solving problems with non linear constraints and thus we have chosen to apply this optimization technique on our new optimization problem. The algorithm is written in the Modeling Language for Mathematical Programming (AMPL).…”

Learning of grasp adaptation through experience and tactile sensing

“…However, as studied in [2], [12], [13], the object physical properties that are usually unknown prior to contact, such as coefficient of friction, object weight and center of mass, can also significantly affect the stability or feasibility of the final grasps. For instance, a robotic hand will likely drop a grasped object when lifting it if the object is heavier than assumed, even though the hand is in the desired configuration.…”