Abstract : In this paper, we present an optimization-based retargeting method for precise reproduction of captured human motions by a humanoid robot. We take into account two important aspects of retargeting simultaneously: spatial relationship and robot dynamics model. The former takes care of the spatial relationship between the body parts based on "interaction mesh" to follow the human motion in a natural manner, whereas the latter adapts the resulting motion in such a way that the dynamic constraints such as torque limit or dynamic balance are being satisfied. We have integrated the interaction mesh and the dynamic constraints in a unified optimization framework, which is advantageous for generation of natural motions by a humanoid robot compared to previous work that performs those processes separately. We have validated the basic effectiveness of the proposed method with a sequence of postures converted from captured human data to a humanoid robot.