Neural Network Control of Pneumatic Muscle Robot With CPG Spindle Reflex

The research subject is a 2-degree-of-freedom leg mechanism that utilizes two opposing pneumatic muscle joints connected in sequence. On this basis, a dynamic model and a neural network control law is established, the trajectory tracking and anti-interference research are carried out by integrating the Central Pattern Generator and spindle reflection. Firstly, a neural network adaptive controller is designed for PAM single-leg dynamics model of the PAM quadruped robot. The neural network-estimated joint torque is employed to address the issue of pneumatic muscle pressure disparity, other, the rhythm signal under the walking gait of the Central Pattern Generator is used as the upper control input. Secondly, the rate of change of the contraction velocity of the pneumatic muscle under external disturbance is analyzed, and the external model of the spindle reflection of the Central Pattern Generator is established to compensate for the external disturbance. Finally, simulation verification is carried out in two cases half-sine pulse and trapezoidal pulse. The outcomes of the experiment indicate that the target signal can be efficiently rectified.