Computation of Suitable Grasp Pose for Usage of Objects Based on Predefined Training and Real-time Pose Estimation

Abstract

Existing grasping mechanisms focus on executing accurate grasps which are not always suitable for the usage of objects. We developed a system that can be used to train humanoid robots with different types of grasp poses. We present a grasping mechanism using homogeneous transformation that allows a humanoid robot to grasp objects in such a way that is suitable for the usage of the objects. The system captures the relative poses of an object and a robot's wrist for training such that when the object's pose changes, the robot's gripper attached to the wrist adjusts its pose accordingly and lines up with the object. For detecting the objects and estimating their poses, we developed and used a color-based pose detection and estimation system and a homography-based planar pose detection and estimation system. We conducted experiments using a humanoid PR2 robot. We used the Robot Operating System as the primary framework of the system and MoveIt Interface for manipulation of grasps. The grasping system showed robust results for different poses of the objects using both arms of the robot. Our experiments involved human validation in which the robot successfully grasped objects such as a screwdriver, a wrench and books from human hands in different grasp poses that are appropriate for usage ofthe objects.