Human guided reach and grasp planner

Abstract

Humans are extremely adept at grasping and manipulating a large variety of novel objects. The current state of robotic grasping, though improving, has not approached the level of robustness found in humans. In order to improve robotic grasping, this paper describes a method of integrating human grasp data into automated path planning and grasp acquisition called the Human-Guided Reach and Grasp Planner. Human grasp data is collected by combining input from two hardware systems: an array of infrared motion tracking cameras for spacial orientation and a sensor glove for highly accurate joint information. Collected data is stored for later use in robotic grasp planning. The Human-Guided Reach and Grasp Planner is then able to generate a trajectory for accomplishing the demonstrated task on the same object in new environments. This planner combines previous research in the fields of robotic learning, sequence alignment, human grasping and path planning among others.