2-tier control of a humanoid robot and use of sign language learned by Monte Carlo method

In this research, an approach to implement a collaborative task between a humanoid robot (Hubo) and a human is presented. Velocity control using motion data generated from a motion capture system (MoCap) is used to control Hubo's lower body movement. The difference in moving direction and speed between the robot and a worker produced a step distance and turning angle of subsequent steps. For upper body control of Hubo, passive control enables the robot's arms to respond adaptively to human arm movements and diminishes undesired reaction forces from a human worker. For better interactive collaboration, several messages were chosen to assist communication between a human and Hubo. For each specific message, various kinds of sign language were initially designed and collected by MoCap. Captured signs were evaluated using Monte Carlo method and an optimized sign was determined based on the stability of carried objects and the robot itself. Finally, an experimental evaluation of the presented approach with the chosen signs was demonstrated through a real collaborative task between Hubo and a human worker which was carrying panels of various sizes.