Artificial neural network controllers for biped robot

This paper presents comparison of three artificial neural network controllers design based on cascade-forward, feed-forward neural network and radial basis neural network to control level walking of biped robot. The biped robot consists of a hip, knee and ankle of both legs and torso. It uses the experimental flexion angle data of seven-link movements of human for level walking. The simulation environment contains a model of the robotic leg dynamics and different neural networks for inverse dynamics of leg. It has three independent neural networks of three joints separately in order to achieve the level walking. The simulation work is carried out in Matlab. The results showed that the radial basis neural network is better and can be used to control level walking of a biped robot.