Fuzzy control of dynamic locomotion by using leg-force distributions of quadruped robots

This paper describes a fuzzy control method for stabilizing the trot gait of quadruped robots, where it is assumed that force sensors are attached in each sole to measure the leg-force distributions. For designing the fuzzy controller off-line, a radial basis function (RBF) neural network (NN) is constructed to identify the relation of joint angles and forces of two support legs. After regarding the RBFNN as an actual robot, the design parameters of the fuzzy controller are tuned by genetic algorithm (GA), minimizing a fitness function composed of a square-root error of zero moment position (ZMP). The practical experiments prove that the fuzzy controller can play a role so that the consequent ZMP of the robot is assigned on a diagonal support line. Copyright (C) 2001 IFAC.