Fast and Reliable Contact Computations for Grasp Planning

Grasp planning is a key problem in robotics. One of the goals is to find suitable forces and torques for an object through contact points for a robotic hand to grasp that object. At a broad level, there are two classes of grasp planning algorithms that are used to determine the relationship between the contact points and finger joint positions. The forward methods simulate the motion of the hand and finger closing and involves use of collision detection methods to calculate the contact points. On the other hand, the backward methods first locate the contact points on the object surface and compute the feasible finger joint positions using inverse kinematics on the articulated model of the finger. The main drawback of the backward methods is that collisions between the hand and the environment during the inverse kinematics computation and finding collision-free configurations can be non-trivial. On the other hand, the forward method can easily find grasp configurations without any collisions with the environment. In earlier work, one of the major issues with respect to forward grasp planning methods was finding the contact points for multi-fingered robotic hands, as it was regarded as a time consuming computation and susceptible to robustness issues.