A hierarchical structure for motion planning and learning control of a biped locomotive robot is introduced. At the upper level of the system, the motion of the center of gravity of the robot is simulated by that of an inverted pendulum. This enables us to compute and predict the position of the center of gravity and also the landing position of supporting toe for next steps. In the second level, in order to determine the positions of other joints from the position of the center of gravity, and of the two toes, we have used an approach known as Hopfield model. However, the inverted pendulum model proposed does not exactly reflect on the actual movement of the robot and there may be some errors in the scheme of neurocomputing. To cope with such situations, the reference input for the position of the center of gravity of the robot is compensated by a learning function of a multi-layered neural network. The system provides an autonomous motion planning scheme for the biped locomotive robots, which is able to cope with, to some extent, the change of walking pattern. Simulation results showed the effectiveness of the proposed method.