Applying motion capture system for collaborative tasks with a human

In this paper, a collaborative work between a human worker and an adult-sized humanoid robot (HUBO) is presented. Through the task, HUBO decided its movement just based on captured data from motion capture system (MoCap), which was equipped in provided working environment. Human co-workers did not provide any cues for determining a desired moving velocity or orientation. Whenever human workers changed their walking patterns, corresponding step distance and heading direction of HUBO was calculated from relative position and orientation difference between the robot and the worker. To generate a stable walking movement, we optimized the initial step of HUBO using reinforcement learning whenever walking pattern is changed. For this, we used a Q learning algorithm with inputs from the Mo-Cap system and Zero moment point (ZMP) of HUBO served as a balance criterion. Experimental evaluation of the presented approach demonstrates that HUBO can follow the movement of a human worker and carry an object without helps from humans such as oral speech commands.