A neural network solution for bipedal gait synthesis

An artificial neural network is used to calculate a set of control torques that are used to generate a locomotive gait for a bipedal robot. Several simplifications of the dynamic model are made. The biped is constrained to the sagittal plane, has no knees, and walking appears in the form of `stilt´-like motion. A supervised learning method is used to train a set of two fully connected multilayer feedforward neural networks. Training data from several mathematically derived linear control laws are accumulated into a single training set. The neural network solution is to incorporate and combine information from the results of several linear control methods. The results of these simulations indicate that the neural network approach for generating controlling torques could far outperform teaching controllers