Motion planning for virtual human upper body including torso and arm received continuous interest in computer graphics community in the past year. Though a variety of motion planning approaches have been proposed to address the problem of 7 DOF (degrees of freedom) arm manipulation, planning the motion of entire upper body chain for general manipulation task is remained unresolved as there is no explicit inverse kinematics (IK) solution for the redundant chain. In this paper, a novel unified framework that combines the workspace goal-oriented heuristics and a variant of random sampling strategy called Goal-oriented Rapidly-exploring random tree (Goal_RRT) planner is proposed to efficiently resolve manipulation planning problems for upper body chain without the need of explicit IK solutions. Experimental results demonstrate the method is fast and reliable.