Abstract:The theoretical model is the foundation for the control of pneumatic artificial muscles. The static characteristics of pneumatic artificial muscles are affected by many factors. By considering the main factors, the relationship between the force, pressure and contraction of pneumatic artificial muscles are derived based on the Chou ideal model, and then a modified perfect model is built. On the basis, a neural network cascade ProportionIntegrationDifferential (PID) controller for a joint driven by pneumatic artificial muscles was designed, by taking the advantages of neural network control scheme such as selfadaptability and robustness. The cascade closedloop controller was adopted of an inner pressure controller and an outer position controller to control the pressure and position respectively. Performance effect of the cascade controller is verified by experiments.