Manipulator-based grasping pose selection by means of task-objective optimisation

This paper presents an alternative to inverse kinematics for mobile manipulator grasp pose selection which integrates obstacle avoidance and joint limit checking into the pose selection process. Given the Cartesian coordinates of an object in 3D space and its normal vector, end-effector pose objectives including collision checking and joint limit checks are used to create a series of cost functions based on sigmoid functions. These functions are optimised using Levenberg-Marquardt's algorithm to determine a valid pose for a given object. The proposed method has been shown to extend the workspace of the manipulator, eliminating the need for precomputed grasp sets and post pose selection collision checking and joint limit checks. This method has been successfully used on a 6 DOF manipulator both in simulation and in the real world environment