Postural control of a two degree of freedom biped using a simple neural network

A neural network controller, employing 3 neuron-like adaptive elements, was used to control a planar double-inverted pendulum biped model. The net learned to balance the biped about the vertical equilibrium point and maintain dynamic stability. Simulations show consistent similarities between model dynamics and human postural dynamics, despite the simple controller architecture, and encourage further evaluation using more elaborate models.