A neural gait synthesizer for autonomous biped robots

An autonomous gait synthesis mechanism based on neuro-computing is presented. The mechanism is for generating motion trajectories of biped robots in negotiating difficult terrains. It is centered on a neural gait synthesizer. The latter consists of a number of functional unit including a central pattern generator an adaptive neural network, a knowledge base, a learning unit and a switch mechanism. The central pattern generator is responsible for generating motion patterns of voluntary and involuntary motions; the adaptive network is used to modify the reflexive motion patterns in accordance with terrain conditions; the responsibility of the switching unit is to make decisions in real time to switch between voluntary and involuntary motions; the knowledge base is used to store feature parameters of motion patterns, and the learning unit extracts feature parameters from an involuntary motion. Based on the functional units, an architecture for the automated gait synthesizer is presented