Control of stability in semi-passive robot based on RBF neural network

The stability control of passive walking has been a major challenge in the field of passive robots because of its significant nonlinear characteristics. Using the adaptive sliding mode control based on local approximation of Radial Basis Function Neural Network (RBFNN), the walking stability of the simplest point foot model of biped passive robot is investigated in this paper. Simulation results show that the passive robot with this control method could achieve stable periodic gait with fewer steps than the robot without the control. In addition, given different control target state functions according to the characteristics of passive walking gait, passive robot can stabilize to corresponding periodic walking state with small control errors under the adaptive sliding mode control based on RBFNN. Finally, the influences of the gain coefficients of weight updating formula of RBFNN and the robustness coefficients of adaptive sliding mode controller on the control error results are discussed. It is found that the error decreases when the parameters increase within a certain limit, and the control effect becomes less obvious when the parameter exceeds the limit range.