Robot reaching movement synthesis by human demonstration

Abstract

Reaching to an object is a fundamental skill to a robot. The purpose of robot reaching aims at bringing the robot hands to the object location without obstacle collision. Since implementing such robot reaching movement is a tedious task, this paper proposes a method to synthesize robot reaching movement by human demonstration. However, human demonstrations are inconsistent, this paper adopts Gaussian Mixture Model (GMM) to obtain generalized reaching movement which is then used to synthesize new movement to avoid the obstacle by adjusting the parameters of its Gaussian model. With the proposed method, robot reaching movement is goal-directed, collision-free, and human-like. We validated the proposed method on a Aldebaran Robotics NAO humanoid robot with 25 degrees of freedom. The results showed that the NAO robot was able to perform skillful reaching movement to the object.