How can human motion prediction increase transparency?
A major issue in the field of human-robot interaction for assistance to manipulation is transparency. This basic feature qualifies the capacity for a robot to follow human movements without any human-perceptible resistive forces. In this paper we address the issue of human motion prediction in order to increase the transparency of a robotic manipulator. Our aim is not to predict the motion itself, but to study how this prediction can be used to improve the robot transparency. For this purpose, we have designed a setup for performing basic planar manipulation tasks involving movements that are demanded to the subject and thus easily predictible. Moreover, we have developed a general controller which takes a predicted trajectory (recorded from offline free motion experiments) as an input and feeds the robot motors with a weighted sum of three controllers: torque feedforward, variable stiffness control and force feedback control. Subjects were then asked to perform the same task but with or without the robot assistance (which was not visible to the subject), and with several sets of gains for the controller tuning. First results seems to indicate that when a predictive controller with open loop torque feedforward is used, in conjunction with force-feeback control, the interaction forces are minimized. Therefore, the transparency is increased.