In this paper, a humanoid grinding robot with two 6-DOF manipulators is proposed. The robot is equipped with force/torque sensors and binocular vision cameras, through which the interaction between the robot and the environment can be recognized and serves as a feedback for adaptive grinding control. Kinematics of the dual-arm robot is analyzed. Simulation of a grinding movement is carried out in ADAMS, during which a propeller blade weighing 40 kilograms is moved following desired trajectories. Then joint torques of the two manipulators are obtained. This paper further proposes an architecture of position-force control of the dual-arm robot. The simulation results show that the dual-arm robot performs well in grinding.