Tele-teaching by human demonstration in virtual environment for robotic network system

In this paper we present a system that is able to give force feedback for the handling of virtual objects to a human operator and discuss how this system can learn low-level reflective behaviors from the human. A dynamic force simulator (DFS) that simulates object dynamics, contact model and friction characteristics of the human hand was developed to interact with the object in virtual reality. The DFS allows calculation and feedback of appropriate forces to the force controlled actuators of a glove type haptic interface. The measured data is used for task-teaching of the robot in the remote site. However usually the task can not be achieved by simple playback teaching. As each task execution is different due to various sources of small errors a low-level motion behavior to compensate for this errors is needed. In this paper, this function is assumed to be a reflective behavior and we acquire it by learning from human. Two learning schemes, neural nets and radial basis functions are experimentally evaluated