Upper limbs of human beings are extremely special and significant, which obtain crucial function to achieve manipulations intelligently and dexterously. Gesture-Changeable Under-Actuated (GCUA) grasping function is presented to improve the capability of robotic hands to achieve humanoid manipulations with low dependence on control and sensor feedback systems, which includes traditional under-actuated (UA) grasping motion and special prebending (PB) motion. Based on GCUA function, GCUA Hand II is developed, which has 5 fingers and 14 DOF. All the fingers use similar tendon mechanisms and motors to achieve GCUA function. With GCUA Hand II, a humanoid robot upper limb system is designed, which has two 3-DOF arms actuated by stepper motors and two 14-DOF hands actuated by DC motors. The control system includes four parts: a computer, a FPGA motion controller, a driver module, and a user module, which can control the upper limb system to do various movements dexterously and exactly. With C++ language, a spatial motion program is designed to assist researchers to determine spatial motions of the upper limb system. This system has a great prospect in the field of rehabilitation engineering, extremely environmental manipulation, humanoid robotics and social services.