For solving the problems which exist when building correct mathematical models for strong coupling, nonlinear, time-changing systems such as robot operating arms, and these problems often make traditional control methods have not good enough decoupling results and bad dynamic performance, therefore we put forward one improved RBF neural network inverse model based decoupling control method. Firstly, the particle swarm optimization algorithm is utilized by us to carry out offline optimization for initial neighborhood clustering radii, therefore obtaining relatively better network parameters. These optimized radius values are then employed for real-time nearby region grouping to dynamically build the RBF neural network reverse model, thus overcoming the restriction of traditional methods which need pre-set network structures. Second, the inverse model that has been found is connected in series with the system being controlled to construct a pseudo-linear system, therefore it enables dynamic decoupling between the joints of robotic manipulators. In the end, one PD controller is combined into a compound closed-loop control system for reducing inverse model errors and strengthening robustness. The simulation which uses a two-degree-of-freedom manipulator to carry out validation proves that our strategy can effectively realize dynamic decoupling and high-precision track following, whose tracking accuracy and robustness all exceed the traditional PD control methods.
