Endowing the robot with a certain amount of cooperation experience before tasks is essential to realize its active compliance control. Through analyzing the process of shaping the human experience and observing judgment strategies of an organism, a new intention estimation method for the collaborative robot was proposed. After offline training based on radial basis function BP neural network, the robot could obtain some cooperative skills by the prediction model. During the online execution, the robot could estimate the cooperators intention according to the force information from human. The method can overcome the question in establishing the model of the human robot cooperation based on traditional methods，especially for the complex and dynamic realtime movement model of human and the uncertainty impedance parameters values. Experiment results show that the cooperator’ force is reduced while human robot synchronism motion is developed, so that the compliance of the robot is improved greatly.