Cnn Cpg Theory And The Robot's Gait Controller

The study of the control strategy based on Central Pattern Generator (CPG) has been an international active field in robot's locomotion control. The strategy can be achieved by Cellular Neural Networks (CNN) and main ideas of the process are addressed in this paper. And then, Ensemble Genetic Algorithms (EGA) is proposed and is applied to the parameter-pattern design in CNN. For difficulties in academic analysis on state-equations, a new CNN is built by using a sigmoid function as the networks' output instead of the traditional one. Existence of periodic solutions in the new CNN is proved, utilizing relative stability theories in differential equations and numerical calculation. An approach based on local bifurcation is also introduced to find the critical value when the period solutions vanish. Moreover, a suitable periodic solution can be achieved by changing the value of the bias of the CNN cells, which is an academic foundation to generate different patterns in CPG strategy. Simulation experiments show that all findings in the article are valid. After that, a new CPG model based on CNN and fuzzy control is brought out for shortages of stratagems now available, and Fuzzy Neural Networks (FNN) is employed to conform errors in controlling process. The stability of "autowaves" in CNN is defined and an evaluation function is also introduced to work during the process of building the controller. Genetic Algorithms are also used to offer a lot of reliable training data for FNN. Simulation data show that, the controller has ideal effects, self-adaptability and powerful antijamming capability. Moreover, it is able to be easily achieved by analog circuits.