Grip Synthesis of Three-Fingered Robotic Hand for Unknown Objects

In this thesis, we proposed an algorithm of grip synthesis for multi-fingered robotic hand on arbitrary unknown objects. The synthesis-process was considered as two separate problems: where the robotic hand's fingers should be placed on object with no preconception and how the robotic hand should be configured under its kinematics constraints. The first problem is also known as the synthesis of grasp and the latter one is the synthesis of hand configuration. To find good grasps on arbitrary object, the Genetic Algorithm was taken advantage of in our proposed algorithm for dealing with a large number of possible combinations. The simulated results showed that the group of individuals was able to autonomously evolve to better direction as how it was supposed to. The problem of hand configuration synthesis aims to figure out how the robotic hand should be configured for a given grasp. A method of searching valid hand configurations was proposed and a set of quality criteria were used to estimate the potential stability of candidates. This was tested in simulation and the results indicated that it was feasible and valid for different objects. Finally, a practical testing experiment was carried out with the multi-fingered BH-8 robotic hand. It was proved that the grips synthesized by the proposed algorithm were able to grasp and lift up objects in most cases and have the ability of resisting external disturbance to some extent.