Quadrupedal walking using trained and untrained neural models

A comparative study of three basic approaches to gait control is presented: a simple centrally controlled, state sequenced approach; local reflexive neural clusters modulated by a loosely coupled oscillator; and local reflexes modulated by training algorithms. The training techniques used include a `genetic type´ approach, a temporal difference learning method, and a simple for best return search. Some discussion of theoretical models is given along with a description of an empirical study carried out on a small four-legged test platform