In order to use humanoid robots as a substitute for the human, we have to be able to generate any kinds of humanoid robots' motion. However generating motion for humanoid robots is complicated because we have to generate as many input variables as the number of their joints. Thus in many cases, human motion data is used to generate humanoid robots' motion. However, since capturing all kinds of human motion is not practical, we need to generate various motions by interpolating finite number of motion data. In addition, typically generated motion has to fulfill some kinds of constraint conditions with end effectors. In this paper we use smoothing added constraint conditions as a penalty and Functional PCA to implement motion blending which enables to generate intermediate motion of some motion data.