An Approach to Generate Different CPG Patterns

In this paper a new approach is proposed to generate different patterns for legged robots locomotion control, which is performed by cellular neural networks (CNN) playing the role of artificial central pattern generator (CPG). An analog system is introduced by replacing the output with a sigmoid function to enable some academic research feasible in CNN state equations. A limit cycle is firstly proved existent applying Poincare-Bendixson theorem and numerical calculation in the new system. Critical values of biases are also figured out when the limit cycle disappears and a local bifurcation occurs. And then, a conclusion is derived to show that suitable patterns can be achieved by modifying the values of biases in CNN system, which is a foundation to generate different patterns in CPG control strategy. Simulation results are given to illustrate the suitability of the proposed approach