Acquisition and Realization of a Rolling-Over Motion for a Humanoid with Soft Sensor Flesh

Behavior design with rich sensor information is very important in realizing close interaction using whole-body contact. Our research target is to construct behavior design method taking into account the mutual interaction between a robot’s tactile sensor information and its motion trajectories. In this paper, a behavior acquisition framework using an evolutionary technique is constructed based on a dynamics simulation model including the emulation of tactile sensing. Furthermore, acquired rolling-over motions are tested using an actual humanoid “macra” with soft sensor flesh, and the results are analyzed and discussed for confirming the feasibility of our proposed system.