Sliding Mode Control of 5-link Biped Robot Using Wavelet Neural Network

As a biped robot has become more anthropomorphic and performs a various task on behalf of human, the research on biped robots gradually attracts much attention and is progressing dynamically. However, biped robots are difficult to control due to their nonlinear and coupled dynamics. First, Inverted pendulum [1] is applied to interpret some characteristics of human walking. Later, researchers construct a 3-link biped robot model [2], and a 5-link biped robot model [3,4]. In controlling biped robots, we face some problems such as instability of locomotion, high-order dynamic equation, existence of different phases of the walking cycle and various uncertainties. Due to these constraints, a biped robot requires a robust control technique having higher performance in spite of uncertainties comparing with standard PD control. So, a computed torque or inverse dynamics technique using feedback linearization [5,6] is proposed to control a biped robot. However, such methods are difficult to control a biped robot model with the model uncertainties. Therefore, the sliding mode technique for the robust control of a biped robot with uncertainties is proposed [7]. However, the sliding mode control (SMC) requires prior knowledge of the mathematical model and uncertainty bounds. On the other hand, recently, wavelet neural networks (WNNs), which combine the capability of neural network [8-9] for learning from processes and the wavelet decomposition [10], are used as good estimation tools for the identification and control of dynamic system [11]. Training algorithm plays important role for WNN approximation. The gradient-descent (GD) method is used as conventional on-line training technique. However, the GD method is difficult to acquire sensitivity information for unknown or highly nonlinear dynamics and has the problem, which settles to the local minimum. So, training methodology, which is induced by Lyapunov stability theorem [12], has researched to ensure the stability, robustness, and performance of system. In this paper, we propose WNN based SMC (WNNSMC) for the stable walking of 5-link biped robot with uncertainties. In our control system, wavelet neural network is employed to estimate uncertain and nonlinear functions of the 5-link biped robot. All weights of WNN are trained by the adaptation laws induced from the Lyapunov stability theorem, which are used to guarantee the stability of control system. Finally, in order to verify the effectiveness and robustness of the proposed control technique, the performance of control scheme are proved by comparing the tracking performance of the WNNSMC with that of the SMC via the computer simulations. Sliding Mode Control of 5-link Biped Robot Using Wavelet Neural Network