Grasp generation for assembly using rrt based algorithm

Abstract

When grasping a object with robot arm, hand pose to grasp is one of the important

things to be chosen. This paper proposes an algorithm, Feasible Square Grasp

RRT, which gives an hand pose for grasping and assembling SOMA puzzle with

respect to constraints and collision-free path to the hand pose. There were related

works to integrate grasp planning and collision-free path, but they does not consider

how to grasp regarding usage of object to be grabbed. In this work, constraints are

related to assembly process of SOMA puzzle. First, A set of feasible squares is de-

ned for a piece of SOMA puzzle using which faces of the puzzle are exposed. Grasp

poses are computed using the feasible squares. Second, the algorithm grows a tree

randomly and nds collision-free path of a robot arm from arbitrary given initial

pose to a feasible grasp pose using inverse kinematics method. Finally, path from

grasp pose to assembly pose is computed using the tree. The algorithm goes back

to feasible square part and iterate for each pieces of SOMA puzzle. Feasible Square

Grasp RRT is evaluated with simulation by assembling a 3 by 3 cube with SOMA

puzzle, using 6 DoF robot arm with parallel gripper.