Dexterous Humanoid Whole-Body Manipulation by Pivoting

Recent progress in research on humanoid robots is making them to complicated tasks, such as manipulation, navigation in dynamic environments, or serving tasks. One of promising application areas for humanoid robots include the manipulation task thanks to their high potential ability of executing of a variety of tasks by fully exploiting their high mobility and adaptability coming from its large number of degrees of freedom. Especially, manipulating bulky objects through a whole-body motion is suitable for them, which has been difficult for other types of robots. This paper focuses on whole-body manipulation of large objects by a humanoid robot using a method called "pivoting" (Y. Aiyama, et al, 1993). This manipulation method has several advantages such as dexterity and stability over other methods like pushing or lifting. Moreover, the requirement of maintaining the equilibrium of the robot during the manipulation cannot be managed in the same way as in the case of the robot simply walking. To cope with those problems, an impedance control framework is first introduced to hold and manipulate the object, together with a whole-body balancing control. Next, a control framework called resolved momentum control (RMC) (S. Kajita et al., 2003) is adopted to allow the robot to step forward after manipulation by keeping the hand position with the object. The next section presents overview of the subject of manipulation tasks. Section 3 addresses an algorithm to deal with the manipulation, followed by the description of control techniques included in the algorithm in Section 4. Section 5 gives simulation results using the dynamic simulator OpenHRP. In Section 6 the simulated results of manipulation are verified by hardware experiments using HRP-2 humanoid platform described before concluding the paper.