Neural Network-Based Kinetic Parameter Identification for a Wheel-legged Robot

This paper proposes a neural network-based parameter identification approach to compensate for the modeling errors of the wheel-legged robot. The dynamic model of the wheel-legged robot is established based on the wheel combined variable-length inverted pendulum model. Then, backpropagation neural network is used to identify the dynamic parameters of the wheel-legged robot. Finally, contrast experiments are conducted to verified performance of the proposed approach. Experimental results show that the proposed approach can enhanced the performance of wheeled-legged robot LQR control.