Kinematic control and coordination of walking machine motion using neural networks

An algorithm using neural networks is proposed to coordinate and control the leg movements of a walking machine. The networks are based on the theory of the Cerebellum Model Arithmetic Computer (CMAC), which is a neuromuscular control system. The authors propose an extended CMAC (E-CMAC) to learn the multivariable, nonlinear relationships of the leg kinematics. The E-CMAC networks are applied to perform feedforward kinematic control of a four-legged walking machine in straight-line walking and step climbing. Training and execution of the networks are fast enough for real-time applications and the memory requirements can be easily met. The proposed approach can also be applied to the control of multiple industrial robots.<>