Abstract: A new method is proposed for the motion planning problem of the humanoid robot upper body.First,the kinesiology model is built based on the physical structure characteristics of human body with the knowledge of mechanism and robotics. Second,considering different movement characteristics between the waist and the arm, two human performance measures are comed up: joint displacement and discomfort-potential energy that are then combined into a human upper body motion performance index.Third,based on the index,a new motion planning algorithm of humanoid robot upper body is proposed utilizing the gradient projection method. Finally,the feasibility and the rationality of the algorithm are verified through comparasion between the Matlab numerical simulation and experiment of human upper body.