Design of a controller for an autonomous distributed multi-actuator system using genetic methods

In this paper, an autonomous distributed system consisting of multiple actuators loop-connected into a wheel shape was considered. In order to simplify the controller design procedure, genetic programming and genetic algorithms were employed to directly search for optimal torque functions of this multiactuator system of which nonlinearity and multiplicity of actuators make it difficult to determine the optimal torque functions analytically. As multiple torque functions corresponding to actuators in different states need to be obtained, multiple GP processes which share the same fitness value were executed in parallel. Though it is believed that GP is well suited to difficult control problems where no exact solution is known, at the population size of 50, commonly used GP failed to obtain an acceptable solution for this system. Step by step experiments demonstrate how GP with small population size can be used to improve a predetermined solution by using the proposed concepts of partially hinted initial population and common-branch crossover operator. Separately, another GA search method which manipulates directly on the values of the torque tables was employed and the simulation and experimental results of both methods were compared