Abstract: An adaptive neural network control method based on speed observer is proposed for the control of flexible joint space robot without speed feedback. Based on singular perturbation theory, the dynamic model of space manipulator with flexible joints is decomposed into two subsystems，fast and slow subsystems; Using neural network to approximate the unknown nonlinearity in the controller and observer, a speed observer and a controller based on adaptive neural network are designed to dynamically counteract the influence of model uncertainty on the system. Using Taylor linearization method, an on line adaptive learning law of weight, basis function center and width is designed, which improves the control accuracy and does not need the off line learning stage. A controller based on velocity difference is designed to suppress the elastic vibration in the model of the rapidly varying subsystem. Based on Lyapunov stability theory, the uniform ultimate boundedness of the closed loop system is proved. Simulation results show the effectiveness of the proposed control strategy.

Key words: space manipulator, flexible joints, speed observer, neural network, vibration suppression